Glitchfront

Blog

  • Your LG TV Has Been Spying on You

    Your LG TV Has Been Spying on You

    Image by Timofey Radkevich on UnSplash

    I used to believe that the most disturbing thing I could encounter in my living room was my reflection staring back at me from a blank screen at 3 a.m. I was completely wrong. It turns out that my screen may have been taking notes on me the entire time and doing everything from mapping my Wi-Fi network, recording what I say, and generally listening in even after I have stopped talking.

    Before reading this, maybe you can sit down somewhere your LG smart TV can’t see you?

    What Did the Gamers Nexus Investigation Uncover?

    Gamers Nexus, an outlet that is mostly in the business of benchmarking graphics cards (rather than doing deep dives on television firmware), is the one that let the cat out of the bag. The team at Gamers Nexus got a bit of help from Level1 Techs and a handful of independent security researchers, and they spent more than 500 hours and roughly $70,000 meticulously picking apart retail LG OLED sets, and this includes the G5 which was last year’s flagship model. All this was done for a 135-minute investigation that they posted on their YouTube, and of course, it grabbed everyone’s attention.

    Packet captures taken with Wireshark unveiled the TVs actively sweeping their own local networks, taking note of and cataloguing every single phone, laptop, printer, and smartwatch they could feasibly find an IP address for. 

    That’s not all. 

    They also catalogued the names and signal strength of any Wi-Fi networks in the vicinity even when they had never been invited to join to begin with. 

    All of this is not a requirement to play movies.

    Let’s get into what the microphone captures. According to findings from the investigation, voice prompts are converted into plain text and stored in logs on the TV. The mic also stays activated for about ten to fifteen seconds after you finish speaking, which is long enough to eavesdrop on your friend when they complain about the ad break. 

    Let’s say you decide to pull the Ethernet cable. The TV will not stop working, it just queues everything in internal storage and proceeds to upload the backlog the moment a connection is reestablished.

    According to Tom’s Hardware in an article published on the 9th of September, LG has strongly denied these claims and have doubled down on the statement that LG TVs only process voice data when the voice button on the remote control is pressed and held or when a wake word is recognized.

    In July, the company also made similar claims to TechRadar.

    Be that as it may, researchers muted a set’s main mic via the settings menu and STILL managed to pull a clean recording from a second, hidden microphone the mute button never affects. In light of all this, you are free to make your own conclusions about whether or not these denials are sincere, or just damage control. 

    Gamers Nexus also found that even if the TV was used purely as an HDMI monitor for a laptop, the same data collection kept running. Switching input methods does nothing to fix the problem. The research team also unearthed remote code execution vulnerabilities in WebOS that are currently moving through responsible disclosure. This is a very dire downside that could potentially give a stranger a lot more access than a television has any business granting.

    At one point in the same investigation footage uploaded on YouTube, LG Ad Solutions President, Serge Matta, summarized the company’s take on this in eight words: “We own the glass. We own the TV.” And this is an indirect way of telling you that you don’t have much of a say in the matter, and that the screen mounted on your wall is LG’S property first and it being your appliance is a distant second.

    If we are to be honest with ourselves, this is simply the business model. According to The Register, LG Ads Solutions (LG’s advertising arm), has told potential advertisers that it has access to and owns data from over 216 million LG TVs located all around the world, with 49 million of those being in the U.S., with a combined reach of over 363 million addressable devices if we include phones and tablets using the same network.

    The technology behind this thorough collection is Automatic Content Recognition (Live Plus), which fingerprints whatever you have on your screen and feeds the results to advertisers so they can check if their campaigns are actually working.

    Image by Tingey Injury Lawfirm on UnSplash

    Texas seems to have noticed before most of us even got a whiff of this. In May 2026, Attorney General Ken Paxton announced a settlement that required LG to get informed consent before proceeding to collect viewing data through ACR, display an easy-to-understand disclosure explaining what it collects, offer a genuine opt-out, and quit forwarding that data to entities that have connections with the Chinese Communist Party. It’s genuinely a great win, on paper. 

    After Gamers Nexus’ first report went out in July, LG reportedly responded by silently adding a forced arbitration clause to its terms of service. This effectively closed the door on potential class action lawsuits as more and more people started reading up on what they had actually agreed to. 

    Yet another finding from Gamers Nexus is that LG’s Do Not Sell My Personal Information toggle still ships off by default, and this is the exact opt-out issue that the Texas settlement was meant to fix. The same case that led to the settlement did not leave out Samsung, Sony, Hisense, and TCL. 

    Now let’s move on to the apps which are definitely not much cleaner. Security researchers at Spur discovered that more than 42% of apps that are available on LG’s webOS store, even the ordinary-looking ones like a Pac-man clone, ship with software that is capable of turning your TV into a residential proxy, routing your neighbor’s internet traffic through your home connection without your knowledge. Journalist Brian Krebs got LG to confirm the problem, and LG has since said that it is now working with developers to strip the feature out entirely, or alternatively suspend the apps that carry it.

    What Can You Do to Make Sure Your TV is not Eavesdropping?

    Image by Towfiqu Barbhuiya on UnSplash

    In as much as all this is unsettling, none of it requires you to permanently unplug your TV and go live your life off grid, though I do get why this option may seem enticing on some days.

    Here are a few things you can do:

    • Turn off Live Plus. You can do this by going to your Settings -> All Settings -> General -> System -> Additional Settings. Switch the Live Plus toggle off. This has the function of disabling the ACR at the source.
    • Reduce ad tracking and delete what is already stored. Under Settings, Support, Privacy, and Terms, Advertising, turn on the ‘Limit Ad Tracking’ option, then tap on the ‘Delete My Personal Information’ option in the same menu. This will clear everything that LG already has on file.
    • Give the TV its own network. You can put it on a guest or IoT network separate from your phones and laptops. You can also turn off UPnP on your router to block devices from freely discovering each other. 
    • Always update your firmware. The flaws that were flagged by Gamers Nexus are moving through disclosure and patches, so an outdated TV will not be able to receive them automatically. 
    • You can skip the internet altogether. The advice from Gamers Nexus was straight to the point: disconnect your TV and make use of an external streaming device that you can easily inspect and reset whenever you need to. 
    • Be extremely picky about apps. This advice is backed up by what Spur found hiding in seemingly innocent apps.

    A TV that cost a few hundred bucks is still trying to siphon money from you ages after you bought it. Turn off whatever you can and isolate what you cannot turn off. Another piece of advice: reserve the interesting conversations for a room that does not have a G5 mounted to the wall.

  • GPT-6 Astra vs. Claude Fable 5.1: Who Actually Wins, and Where the Hype Gets Ahead of the Data

    GPT-6 Astra vs. Claude Fable 5.1: Who Actually Wins, and Where the Hype Gets Ahead of the Data

    Image by BoliviaInteligente on UnSplash

    GPT-6 Astra made its very noticeable debut on the 3rd of September, with Sam Altman posting an apology for a messy rollout. OpenAI definitely did not bother with trying to be modest either. The statements from Open AI predictably call it the most intelligent and aligned model in the company’s history. The president of OpenAI, Greg Brockman, told reporters that it’s “not unreasonable” to assume that we have entered the AGI era. These are bold words that pique the interest of skeptics like me. So I checked the scoreboard just to be sure.

    Scoreboards have a habit of telling messier and more fascinating stories than press releases, and this one was no exception. To be fair, OpenAI tooting its own horn is justifiable. Astra does leave other models in the dust in a handful of categories, some of which are refreshingly weird. But Fable 5.1 wins when it comes to the independent, vendor-neutral ranking of general intelligence. Let’s talk more about it.

    Where Astra Stands Out: 3D and CAD

    Astra handles professional design software the same way that a human engineer or digital artist would, it doesn’t just generate an unchangeable, flat 3D shape.  It also has the ability to self-correct and run quality control tests. If Astra designs a 3D object and flags some parts as “weird”, it will trigger a test render, identify its own mistake, and rewrite the software code to implement the corrections.

    BenchCAD: a benchmark that has models reconstructing CAD programs from rendered 3D views. Upload a picture, and Astra can reverse-engineer a shape from the picture. Astra scores an impressive 95.9% versus Fable 5.1’s roughly 84-85% according to Vellum, which is a significant double digit gap.

    A point worth noting is that OpenAI does note that the Claude comparison runs used modified evaluation settings, so this particular gap should be treated as directionally true rather than as gospel.

    Astra also posted a big jump on AutomationBench (41.4% vs. 31.4% for Fable 5.1) and ScreenSpot-Pro, a test of clicking the exact right pixel in a cluttered UI, where it hit 92.7% against Fable 5’s 87.3% (DataCamp). 

    In layman’s terms: if your job involves wrangling desktop software, CAD files, or UIs, Astra would make the perfect assistant.

    The Coding Comparison: Closer Than the Highlight Reel Suggests

    Fortunately (or unfortunately depending on where you stand), there is no SWE-bench bloodbath. This cycle, the benchmark in use is DeepSWE v1.1, a 113-task agentic coding test, in addition to Terminal-Bench 4.0 and an aggregate Coding Agent Index. On DeepSWE, Astra  scores 74.1% compared to Fable’s 67.4% in OpenAI’s own table. 

    A side note worth adding is that independent trackers put Claude Opus 5 73.7% and Muse Spark 1.3 from Meta at 75.4% for the same test. So we can conclude that Astra’s lead in general in this aspect is more of a rounding error, and not a definite knockout.

    Where Astra clearly pulls ahead is Terminal-Bench 4.0, which is the test for the longer and messier multi-step terminal work. Astra scored 58% compared to Fable 5.1’s 55.8%. Of course, the win is modest, but Astra does legitimately seem to be better when it comes to not randomly losing its train of thought part-way through a particularly complex shell session.

    Image by Ilya Pavlov on UnSplash

    But let’s zoom out a bit and look at Artificial Analysis’ Independent Coding Agent Index which combines a number of these tests, and Fable 5.1 pulls ahead and leads 70 to 67.

    Where Astra gains full bragging rights is the efficiency, not the raw score. On efficiency, Astra easily matches Fable 5.1’s performance in coding at less than half of the token cost. A pitch that leads with cheaper, not smarter, is still a decent pitch, but it’s just not the one Open AI highlighted or led with.

    Artificial Analysis’Intelligence Index (which happens to be the broadest and most cited yardstick in the industry and notably not run by either competitor) has recorded Fable 5.1 beating Astra 65.7 to 61.2. Fable 5.1 also wins Humanity’s Last Exam with a very clear margin at 65.0% to 57.2% according to Analytics Vidhya. In light of all this, the claim about Astra being the “world’s most intelligent model” is simply more about OpenAI’s own tables and less about the claim being something that the entire industry agrees on.

    A Note on the Music and Art Claims

    There is a likelihood that you may have bumped into some viral posts claiming that Astra aced a “Bach Benchmark,” reconstructed entire cities in Unreal Engine, or built a 3D animated pianist. This sounded phenomenal, so I looked into it. Unfortunately, I could not reliably verify these claims against any actual benchmark organisation or vendor source. 

    One of the viral threads had to add a correction noting that a widely shared “Astra gameplay” clip was actually an AI-generated video with watermarks from an unrelated tool, not a real model output. Given that, I’m leaving those claims out rather than repeating unverified social-media demos as fact. I also could not find any published benchmark, from any organization, testing either model on “painting people from pictures” — that appears to be internet speculation rather than a real evaluation, so it’s excluded here too.

    The Stuff Only Astra Currently Does

    A few genuinely distinct capabilities showed up in the launch data:

    • Cybersecurity, for better and worse.  Astra is the first OpenAI model to cross the “Critical” threshold on OpenAI’s own Preparedness Framework for cyber capability, solving 88% of reverse-engineering tasks on SRE-Bench in a single attempt (OpenAI). That’s a genuine capability leap, and it’s exactly why access to the advanced version is gated behind a vetted program rather than handed out freely.
    • Math that’s actually saturating benchmarks.  97.6% on FrontierMath Tier 4, a test built specifically to resist AI, versus 87.8% for Fable 5.1 (DataCamp).
    • Half the hallucinations.  Astra’s hallucination rate on Artificial Analysis’s knowledge benchmark dropped from 92% to 51% at maximum effort — a real improvement, though a 51% rate is still nothing to hand your homework to unsupervised (Artificial Analysis).

    So, Does Astra “Outdo” Fable?

    In the categories OpenAI chose to spotlight (CAD reconstruction, terminal work, math, and cybersecurity), the answer is clearly yes. On the benchmark built by neither company to be the fairest overall measure of general intelligence, Fable 5.1 wins, and it isn’t close. Astra’s real headline isn’t “smarter than everything,” it’s “specialized, efficient, and occasionally overpriced for the gains it delivers.”

    Worth remembering, too: Astra costs 2.5x more per token than its own predecessor, and independent labs like Artificial Analysis explicitly frame their numbers as reference points, not purchase decisions (Emergent). I would recommend that you select a model based on workload rather than based on a press release

  • How to Test an API Without Writing a Single Line of Code (Using Postman)

    How to Test an API Without Writing a Single Line of Code (Using Postman)

    APIs are everywhere. Every time you log in with Google, check the weather on your phone, or pay for something online, an API is doing the heavy lifting. They are the connective tissue of modern software, and knowing how to test one is fast becoming a baseline technical skill, not just a developer specialty.

    The problem is that most API testing resources assume you are comfortable writing code and that’s not always the case. Postman is an API platform used by more than 40 million developers and 500,000 organizations worldwide, including 98% of the Fortune 500. Its appeal is simple: it lets you send requests to APIs, inspect responses, and run tests, all through a graphical interface. No terminal. No code. No setup headaches.

    This guide walks you through the full process, from installation to running your first real test.

    What Is an API, and Why Does Testing It Matter?

    An API (Application Programming Interface) defines how two software systems communicate. When you call an API endpoint, you send a structured HTTP request to a server, and it sends back a structured response, usually in JSON format.

    Testing an API means verifying if it behaves as it should: that it returns the correct data, the right status codes, and the right errors in cases where it’s fed bad input. It’s much cheaper to catch bugs at the API layer rather than catching them in production, or in a worst case scenario, after a user has already hit them. 

    Step 1: Install Postman

    Download the Postman desktop app for Windows, macOS, or Linux here. You can opt for the browser version, but I would recommend the desktop app for more stability. 

    You don’t have to go for the free account, but it is recommended. It allows you to sync your work across a number of devices and you can collaborate with your team. Once logged in, you will find yourself in the workspace view, which is where all requests, collections, and environments live.

    Step 2: Understand the Interface

    The Postman layout is built around a few key areas:

    • The address bar: where you enter the API endpoint URL
    • The method dropdown: where you select GET, POST, PUT, DELETE, or other HTTP methods
    • The tabs below the URL bar: Params, Authorization, Headers, Body, and Scripts
    • The response panel: where you see what the server sends back

    You will spend most of your time in the address bar, the method dropdown, and the response panel. The other tabs come into play for more complex requests.

    Step 3: Send Your First Request

    Before testing anything real, try a public API that requires no authentication. JSONPlaceholder is a free, open REST API built for exactly this purpose.

    Here is how to make your first request:

    You will get a response back almost instantly. The response panel will show you:

    • The status code (200 OK means success)
    • The response time in milliseconds
    • The response body in JSON format, with a post ID, title, and body text

    That is a live API call with zero code written. If you want to explore more public APIs to practice on, The Public APIs repository on GitHub maintains a curated list of free, open APIs across dozens of categories.

    Step 4: Read the Response

    The response panel is where the real information lives. Here is what to look at:

    Status Codes

    HTTP status codes tell you whether a request succeeded or failed. According to Postman’s documentation, the most common ones are:

    • 200 OK:  The request worked and data was returned
    • 201 Created: A resource was created successfully (typically from a POST)
    • 400 Bad Request: Your request had invalid syntax or missing parameters
    • 401 Unauthorized: Authentication is required or failed
    • 404 Not Found: The resource you requested does not exist
    • 500 Internal Server Error: Something broke on the server’s end

    Response Body

    The body tab shows the actual data returned. Set it to Pretty for readable, indented JSON. Raw gives you the unformatted string. Preview renders it as HTML if the response is a web page.

    Response Time and Size

    Postman displays both in the top right of the response panel. Hover over the time value to see a breakdown: DNS lookup, connect time, and server processing time. These numbers matter when you are checking API performance.

    Step 5: Handle Authentication

    Most production APIs require some form of authentication. Postman handles the three most common types cleanly, without you writing a single auth header manually. As Postman’s blog explains:

    API Key

    Go to the Authorization tab, select API Key from the dropdown, then enter the key name and value. Postman adds it to the request header automatically. Commonly used for services like OpenWeatherMap, News API, and similar.

    Bearer Token

    Select Bearer Token in the Authorization tab and paste your token. Postman adds Authorization: Bearer <your-token> to the header for you. Bearer tokens are the standard for OAuth 2.0 flows and JWT-based authentication.

    Basic Authentication

    Enter a username and password. Postman encodes them as Base64 and sends them in the Authorization header, which is what the server expects.

    If you are testing an API that requires OAuth 2.0, Postman can handle the full token exchange under the Authorization tab as well. Select OAuth 2.0, click Get New Access Token, fill in the values from the API docs, and Postman retrieves the token and attaches it automatically.

    Step 6: Test POST, PUT, and DELETE Requests

    GET is the simplest HTTP method because it only reads data. The others write, update, or delete it, and they require a request body.

    POST (Create a resource)

    Change the method dropdown to POST. Click the Body tab, select raw, and set the format to JSON. Then type your JSON payload. Using JSONPlaceholder:

    { “title”: “My first post”, “body”: “Testing the API”, “userId”: 1 }

    Hit Send. A 201 Created response means it worked.

    PUT (Update a resource)

    Change the URL to https://jsonplaceholder.typicode.com/posts/1 and the method to PUT. Provide the full updated object in the body. A 200 OK response confirms the update.

    DELETE (Remove a resource)

    Change the method to DELETE. No body needed. A 200 OK or 204 No Content response means the deletion was accepted.

    JSONPlaceholder does not actually modify data; it simulates responses. For testing real write operations, you will need a staging environment or a sandbox API.

    Step 7: Organize with Collections

    Once you have more than a handful of requests, Collections become essential. A Collection is a folder that groups related API requests together. Think of it as a project folder for your API work.

    To create one:

    • Click New in the sidebar
    • Select Collection and give it a name
    • Save any request to the collection by clicking Save > Save to Collection

    Collections are also what you share with teammates, import from API documentation, or run in batch using the Collection Runner.

    Step 8: Use Environments and Variables

    If you are testing across development, staging, and production environments, typing the base URL every time is tedious and error-prone. Environments solve this.

    An Environment is a set of key-value pairs (variables) that Postman injects into your requests. You use them like this in a URL:

    {{baseUrl}}/users/{{userId}}

    Switch the active environment from dev to prod in one click, and every request updates automatically. No find-and-replace. No typos.

    Variable scopes in Postman, from broadest to narrowest:

    • Global: accessible across all workspaces
    • Collection: accessible only within a specific collection, environment-independent
    • Environment: switches with the active environment (dev/staging/prod)
    • Local: temporary, scoped to a single request run

    When the same variable name exists at multiple scopes, the narrower scope wins.

    Step 9: Write Basic Tests (Still No Code Required)

    Postman lets you add post-response checks to any request. Open the Scripts tab, then click Post-response. While these checks technically use JavaScript, you do not need to write any. Click Snippets at the lower right of the code editor to see a list of pre-built checks you can insert with a single click.

    Useful snippets Postman provides out of the box:

    • Status code: Code is 200
    • Response time is less than 200ms
    • Response body: Contains string
    • Response body: JSON value check

    Select any snippet and the code drops straight into the editor. Hit Send, and the Test Results tab in the response section shows pass or fail. That is automated validation without writing a line yourself.

    Common Mistakes to Avoid

    • Sending a POST request without setting the Content-Type header to application/json when your body is JSON
    • Forgetting to include the Bearer prefix before a token when adding it manually to a header
    • Testing against production when a sandbox or staging environment is available
    • Ignoring response headers, which often contain rate limit information, pagination details, and cache directives
    • Not saving requests to a Collection, which means rebuilding them from scratch every session

    Where to Go Next

    Once you are comfortable with manual requests, two Postman features are worth exploring: the Collection Runner, which runs all requests in a collection sequentially, and Monitors, which schedule collections to run at set intervals and alert you when something breaks.

    If you want to go deeper into API fundamentals before testing more complex flows, MDN Web Docs on HTTP is the most reliable free reference for status codes, methods, and headers. Postman’s own Learning Center is also thorough, and entirely free.

    API testing used to mean spinning up a development environment, writing request code, parsing the response, and then debugging why nothing printed to the console. Postman collapsed that process into a few clicks. The interface is not just a convenience for non-developers; most developers use it too, precisely because it removes friction from work that should be fast.

    Start with JSONPlaceholder. Learn what the status codes mean. Save your requests. Then, when you hit a real API in a real project, you will already know what you are doing.

    Featured image: Walkator

  • The Algorithm Did It: Ethical Problems With Using AI to Predict Crime

    The Algorithm Did It: Ethical Problems With Using AI to Predict Crime

    Featured Image: Etienne Girardet

    Picture this. You haven’t broken the law in any way, you live in a certain area and you’re friends with certain people. This life you have and these particular circumstances you’re in are something that an algorithm finds statistically worth flagging. 

    Before you’ve done so much as jaywalk or double park, the algorithm has decided that you’re a person of interest. 

    This is the kind of reality that predictive policing brings into the picture. The future is pre-judged and the concept of presuming one innocent is a quaint relic that the software wasn’t trained on. 

    Predictive policing is the use of AI and data analysis to forecast criminal activity before it happens, or identifying individuals that are seen as likely to break the law. Law enforcement across different countries has been quite receptive to these tools. 

    Despite the very enthusiastic response from police departments, the ethical problems posed by this technology seem to be multiplying faster than the actual crime rates it was meant to reduce. 

    What Does “Garbage In, Garbage Out” Mean in This Context?

    These algorithms are trained on historical crime data, and unsurprisingly, historical crime data is not a neutral record of reality. It just reflects who got policed. 

    The OHCHR’s Special Rapporteur on racism found that predictive policing tools that are based on location create a feedback loop. Officers over-police certain neighborhoods which leads to them recording new offenses there. The algorithm then doubles down and produces predictions that are increasingly skewed and target the same neighborhoods. 

    Past bias inevitably leads to future bias. 

    The NAACP has detailed how reliance on historical criminal data is compromised from the get go because black communities or poor communities have been the targets of disproportionate over-policing for decades. Using the same data to train AI doesn’t legitimize it. 

    There is an additional part that people do not like talking about: sometimes the data is deliberately distorted. Research published on arXiv noted cases in the U.S. (Louisiana and Connecticut) where police officers deliberately lied about the race of people that were stopped or cited for the purposes of covering up discriminatory behavior. 

    New York’s Stop-and-Frisk Problem, Now With Algorithms

    Research published in 2025 found that the highest volume of stop-and-frisk scenarios since the year 2014 were recorded in 2024. Nine out of every ten people that were intercepted were black or Latino. Over 69% of those stops were just dead ends that didn’t reduce crime. All they did was inconvenience certain groups of people. 

    The Software Doesn’t Even Work Well

    The bias issue might have been worth tolerating if the technology was actually extremely accurate, but it’s not. The Markup’s investigation into PredPol (now Geolitica) assessed more than 23,000 predictions that were generated for the Plainfield, New Jersey Police Department. The success rate was under 0.5% with robbery and assault predictions hitting 0.6% and burglary predictions managing a barely-there 0.1%. 

    The company went out of business in 2023. 

    The Global Issue

    The same ethical failures we just highlighted are happening across Europe, Asia, and Africa. In some cases, those failures are amplified. 

    The Ministry of Justice in Britain is working on something that most describe as a murder prediction tool. 

    How does it work? 

    According to freedom of information disclosures obtained by civil campaigners, the tool will tap into data about individuals’ addiction records, mental health histories, and recorded incidents of self harm. I think we all agree that this type of information has no valid place in a pre-crime calculus. 

    India’s Maharashtra state, on the other hand, has been taking strides to expand its MARVEL system. The system shares data collected by the police with a state-owned AI firm and a private company for the purposes of autogenerating case files and enhancing intelligence. 

    The Carnegie Endowment for International Peace has noted that authoritarian governments in the Gulf, East Asia, and Central Asia are the number one adopters of AI surveillance tools. It’s also interesting to note that liberal democracies in Europe have also been racing to adopt their own predictive policing and facial recognition systems with way less scrutiny than the topic deserves. 

    Africa’s case is an example of what happens when surveillance technology arrives in the absence of an adequate legal framework to govern it. 

    Under the banner of “safe city” and “smart city” programmes, governments across the continent have been deploying AI-powered surveillance systems, primarily financed and built through partnerships with Chinese firms. Huawei’s Safe Cities project furnished Nairobi and Mombasa with 2,000 CCTV cameras; Uganda’s Kampala received a $126 million system with 1,800 cameras and facial recognition capability; Zimbabwe’s Zim Cyber City project involves Chinese firms including Hikvision and CloudWalk Technology.

    A 2025 investigation found that these smart city projects, presented publicly as crime-fighting infrastructure, have in practice been used to surveil journalists, opposition figures, and political opponents, with spyware like Pegasus allegedly deployed during election cycles in Kenya. 

    In Nairobi, crime rates reportedly increased after the Safe City project was implemented, which is a detail that tends not to feature in the vendor brochures. ENACT Africa has noted that across Africa, private security companies are typically far more technologically advanced than the police forces they operate alongside, many of which lack basic internet access and data infrastructure. Into that gap come foreign vendors, usually without democratic oversight or independent evaluation. 

    The continent’s weak or absent data protection frameworks mean that algorithmic systems trained on corrupted, colonial-era policing patterns can be deployed at scale, with no legal mechanism to challenge their outputs. Africa imports the technology and absorbs the consequences, while the intellectual property, the profits, and the accountability remain elsewhere.

    Transparency? What Transparency?

    One particularly frustrating feature of the predictive policing industry is its opacity. These are proprietary systems: private contractors sell black-box tools to public agencies, and neither the affected communities nor their elected representatives have meaningful access to how decisions are made. The NAACP flagged this directly, noting that the proprietary nature of these algorithms prevents any public understanding of the decision-making framework. You can be surveilled, flagged, and policed by a formula that no one outside a corporate legal team has read.

    The Electronic Frontier Foundation has pointed out that the predictive policing market has been consolidating around a small number of vendors, with SoundThinking (formerly ShotSpotter) having already absorbed both Hunchlab and Geolitica.

    When harmful and flawed technologies bundle together, municipalities that buy one product often end up deploying a full suite, without meaningful elected oversight or public accountability.

    The Presumption of Innocence, Quietly Retired

    Maybe the sharpest ethical edge here is the one that gets the least attention: the philosophical rupture at the heart of predictive policing. Western legal tradition is built on the idea that you are innocent until proven guilty. Predictive policing inverts this. It treats the probability of future guilt as sufficient justification for present-day police attention. The ACLU has described this as creating a presumption of guilt by association, in which individuals and entire neighborhoods are flagged based on statistical inference rather than observed behavior.

    Predictive policing’s person-based variant makes this even more explicit: a score is assigned to an individual, not a place, and that score shapes how police interact with them. Chicago’s Strategic Subject List reportedly assigned scores to 56% of Black men aged 20 to 29 in the city, according to a Chicago Magazine analysis. Not because of what they had done, but because of where they lived, who they knew, and what the algorithm decided those factors meant. That is not law enforcement. That is bureaucratic pre-crime.

    What Would Responsible Use Look Like?

    Not everyone calls for an outright ban, though the EFF and several cities have done exactly that. More moderate voices call for independent oversight of algorithms, mandatory disclosure when these tools are in use, and audit mechanisms to review whether outputs are fair and accurate. The NAACP recommends establishing rigorous, independent oversight bodies and requiring law enforcement agencies to disclose not just that they use predictive tools, but how those tools work and what data they consume.

    The legal landscape is also shifting. The Johns Hopkins University Law Review noted that existing equal protection frameworks require proof of discriminatory intent, not just discriminatory impact, making algorithmic bias difficult to challenge in court under current standards. Updating those frameworks to account for statistically demonstrable harm may be the only way to impose meaningful legal accountability. The law, in other words, has a software update pending.

    The Honest Conclusion

    Predictive policing promises science and delivers prejudice with extra steps. The tools don’t work well, the data they train on is structurally compromised, the companies selling them have limited accountability, and the communities absorbing the consequences tend to be the same communities that have been absorbing consequences from the criminal justice system for a very long time. An algorithm that can’t predict burglaries with better than 0.1% accuracy is not a crime prevention tool. It is an expensive way to automate a pre-existing bias.

    If AI is going to play a role in public safety, it will need to be built on clean data, governed by transparent oversight, and subject to meaningful civil rights scrutiny. Until then, the most accurate prediction these systems produce is that the people watching will be the same ones who have always been watched.

  • A Coin-Sized Device Just Changed Everything

    A Coin-Sized Device Just Changed Everything

    For years, we’ve always known the brain interface race to belong to Silicon Valley by default. We have all seen a headline about Elon Musk’s Neuralink. The technology attracted billions in funding and created such an aura of inevitability that we all assumed the first commercially available brain chip would have a Californian zip code. But, on the 13th of March, 2026, a Shanghai startup most people had never heard of rewrote that story. 

    Neuracle Medical Technology released NEO, which just became the first invasive brain-computer interface (BCI) product in history to get full commercial approval from a national regulator. 

    The startup got a proper product license with their product being sold to patients, and already covered by China’s national health insurance.

    NEO is the first invasive brain-computer interface product cleared by any national regulator for commercial use, anywhere in the world. Neuralink, Synchron, and every comparable Western device remain in clinical trials.

    What Is NEO, and How Does It Work?

    NEO is roughly the size of a coin. It was developed jointly by Neuracle Medical Technology (a Shanghai based startup), and researchers at the School of Biomedical Engineering at Tsinghua University in Beijing.

    The procedure to place the chip inside the skull takes roughly ninety minutes. The chip’s eight sensors sit on the dura mater, which is the brain’s outer protective membrane, and they don’t touch the brain itself. All a patient has to do is imagine moving their hand, and NEO will pick up the resulting neural signals and wirelessly relay them to an external computer. 

    The computer then decodes the signals and sends commands to a pneumatic robotic glove the patient wears on their hand. The chip translates intention into movement. A patient can just think of picking up a cup of tea, and the hand closes around a cup. 

    A patient named Dong in MIT Technology Review’s reporting, gave this description on the ninth day of training: “My right hand successfully grabbed a ball without the glove. That was a miraculous moment.” 

    What Did China’s Regulator Actually Approve?

    On the 3rd of March, 2026, China’s National Medical Products Administration (NMPA) approved a Class III medical certificate for NEO, with Class III being the country’s highest regulatory classification. Specific criteria must be followed. The patients must be between the ages of 18 and 60, with quadriplegia caused by cervical spinal cord injuries, and retaining some upper arm function but still unable to grip objects. 

    NEO was used in a total of 36 clinical procedures: four early feasibility trials and 32 multi-center trials and these were conducted under Good Clinical Practice (GCP) standards. 

    Some of the supporting data remains in a preprint that has not yet completed full peer review, but the results are promising 

    Every single one of those 36 patients showed improvements in grasping objects. Some showed signs of neural remodeling, which means that their brains began reorganizing around the device over time.

    NEO vs. Neuralink: The Design Choice That Won the Race

    The single biggest reason NEO cleared regulators ahead of Neuralink likely comes down to one engineering decision: where to put the sensors.

    Neuralink’s N1 chip fires ultra-thin electrodes directly into the cortex, the brain’s outermost layer of neural tissue. The signal quality is potentially superior. But piercing brain tissue means higher surgical risk, greater infection exposure, and the very real problem of scar tissue building up around the electrodes over time, degrading signal quality as years pass. Getting the FDA comfortable with that risk profile has proven to be, as Neuralink discovered in 2022 when the FDA initially rejected its trial application, a slow and methodical process.

    NEO’s sensors sit on the dura mater and that means that they never contact the brain directly. Avinash Singh, a BCI researcher at the University of Technology Sydney, has described this as a significantly less invasive design that contributed to NEO’s faster regulatory path. It is a conscious trade-off: somewhat lower signal resolution in exchange for a much cleaner safety profile and, it turns out, a much faster route to market.

    Mao Ying, president of Huashan Hospital at Fudan University, put it precisely: the system achieves stable acquisition of brain signals through minimally invasive implantation outside the dura mater, without contacting brain tissue or damaging neurons, while accurately decoding patients’ motor intentions. In plain English: all the signal, none of the holes in the brain.

    Where Is Neuralink Right Now?

    To be fair, Neuralink isn’t asleep in a corner somewhere. Scientific American reports that 21 participants were enrolled in its human trial as of January 2026. Its first recipient, Noland Arbaugh, has demonstrated the ability to control a computer cursor using thought alone. Other American startups including Synchron and Paradromics are running their own promising trials.

    But Neuralink is still a trial and NEO is a product. As of March 2026, patients in China can go to a hospital, pay for the procedure, and receive a NEO implant on a commercial basis.

    The FDA is right to be cautious. Piercing someone’s brain cortex carries real risks that should be addressed before a product becomes commercially available. China won the race because they designed around those concerns rather than trying to argue past them.

    Neuralink: 21 trial participants as of January 2026. NEO: full commercial approval, Class III certification, national health insurance coverage, prescriptions available at hospitals.

    China’s Playbook for Brain Technology

    NEO’s science is definitely genuine, but there’s something other than scientific merit that helped the company get a foot through the door. The Chinese government decided, at the highest levels of policy, that brain-computer interfaces are a strategic national priority.

    China’s five-year plan for 2026 to 2030, which was published on the same day Neuracle received its approval, designates BCI as one of six key industries important to the country’s future technology competitiveness, alongside quantum technology and humanoid robots, according to MIT Technology Review. Shanghai alone has cultivated 60 companies dedicated to BCI research and development.

    The Chinese government backed trials and funded the research. It ran the fast-tracked approval pathway, then proceeded to publish the five-year plan the same morning of the approval (in a very calculated move).

    China has done something similar to this before with electric vehicles, solar panels, and more.

    The brain-computer interface market appears to be the next industry where state coordination is being used to leapfrog competitors who are moving at the pace of venture capital and cautious regulators.

    What Does the World’s First Commercial Brain Chip Actually Mean?

    For patients with cervical spinal cord injuries, the immediate practical meaning is substantial. Effective treatments for that condition have been, as Digital Watch Observatory notes, extremely limited for decades. A device that allows someone who cannot grip objects to do so again at home and independently, within a month of surgery, is nothing short of groundbreaking.

    The broader geopolitical meaning is harder to miss. Brain-computer interface technology is not just a medical device market. It is foundational infrastructure for a future in which the boundary between human cognition and machine intelligence gets blurry. The country that builds the dominant clinical infrastructure, manufacturing base, and regulatory framework for this technology first will have an enormous structural advantage as the applications multiply.

    China just took a meaningful lead in that direction. The research team has already signaled its next target: patients paralyzed by cerebral ischemic strokes, with expanded trials planned before the end of 2026.

    The Race Is On

    NEO is not the final chapter of the brain-chip story. It is the chapter where the technology stopped being a research project and became something a patient could ask a doctor about. That shift, from trial to prescription pad, is a bigger deal than any technical benchmark.

    Elon Musk built the narrative around brain-computer interfaces. He deserves credit for that. He turned a niche field of neuroscience into something the general public actually talks about. But Neuracle Technology, backed by Tsinghua researchers and a state that decided to win this race explicitly, built the first device the market is actually allowed to buy.

    For the patients who have spent years not waiting for the most spectacular chip, but for any chip they are allowed to use, that difference is not academic. It is everything.

    Featured Image: Mirella Callage

    Sources

    Nature: China approves brain chip to treat paralysis (March 2026)

    MIT Technology Review: China has approved the world’s first invasive brain-computer chip (June 2026)

    Scientific American: China just approved its first brain implant for commercial use (March 2026)

    Bloomberg: China Approves First Brain Implant for Commercial Use (March 2026)

    Xinhua: China grants world’s first market approval for invasive BCI product (March 2026)

    ZME Science: China Beat Elon Musk’s Neuralink to Approve the World’s First Commercial Brain Implant (March 2026)

    BrainFacts.org: In a First, China Approves Brain Implant for Commercial Use (April 2026)

    People’s Daily: China approves world’s first implantable brain-computer interface (April 2026)

    Digital Watch Observatory: China approves world-first brain chip to treat paralysis (March 2026)

    Greek Reporter: China Approves World’s First Brain Chip to Restore Movement in People With Paralysis (March 2026)

  • Why Is the U.S. So Interested in Africa’s Health Data?

    Why Is the U.S. So Interested in Africa’s Health Data?

    Summary: The United States has snuck in a few health data and pathogen-sharing requirements inside multi-billion-dollar bilateral aid agreements with several African governments. Critics are up in arms and they argue that this type of access gives Washington decades-long, unrestricted access to genomic databases, disease surveillance systems, and biological samples that will feed pharmaceutical pipelines, AI tools, and biocides planning with very little benefit to the people giving the data. 

    Featured Image: Clay Banks

    There’s a sentence that’s buried inside the America First Global Health Strategy and you’d have to dig deep to find it. Craftily hidden among co-financing targets and disease surveillance frameworks is a clause that requires signatory African countries to share sensitive health data and pathogen samples with the U.S.

    And this requirement sometimes stands for up to 25 years. Depending on which side you’re standing on, this is either a simple, reasonable condition for receiving billions in health funding, or it’s the sneakiest, most sophisticated data extraction deal that’s ever been paused to African nations. 

    By the end of February 2026, 18 African countries had taken the bait and signed these bilateral health agreements worth US$19.8 billion, with the involved African nations pledging US$7.5 billion of their own funds. 

    Signatories include Botswana, Nigeria, Kenya, Ethiopia, Uganda, Rwanda, and more. These are countries that collectively host some of the world’s most epidemiologically significant populations and disease landscapes. 

    What Are These Deals Exactly?

    On the 18th of September, 2025, the U.S. government released its new America First Global Health Strategy, and this outlined five-year Memorandums of Understanding (MOUs) with partner countries in the 2026-2030 period. Since USAID has been scrapped with much controversy, this strategy was meant to replace USAID-channelled aid with direct bilateral compacts designed, in Washington’s framing, to help countries build ‘more resilient and durable health systems’ and stand on their own feet without depending on donors. 

    The logic was that the U.S. cuts aid, countries get to co-invest in their own health systems, and both parties get to share data and pathogens quickly. 

    Global health financing dropped 21% from 2024 to 2025 according to the Institute for Health Metrics and Evaluation, so in the context of governments who were already spending far below the 15% set in the 2001 Abuja Declaration, the deals were almost impossible to pass up. 

    Kenya was the first to sign in December 2025 and it agreed to a framework worth approximately US$2.5 billion. But in a twist, within a week, a Kenyan court suspended implementation because of data security concerns. 

    Other governments weren’t discouraged by this and by February 2026, 18 countries had signed and others were open to negotiations. 

    What Does the U.S. Get Out of This?

    Africa’s health data has strategic value that goes beyond disease monitoring and the U.S. knows this. 

    Pathogen samples, genomic databases, and longitudinal electronic health records are inputs for AI tools, pharmaceutical pipelines, and security analytics that generate economic and strategic value far beyond Africa, according to researchers at the Centre for Global Health Law at Warwick University and KELIN, a Nairobi-based health rights organization, writing in PLOS Global Public Health.

    Nelson Aghogho Evaborhene, a PhD Fellow in Global Health Governance at Roskilde University in Denmark, said it quite plainly: pathogen data is what underpins the entirety of the biomedical innovation pipeline. “Countries and companies that had early access to viral sequences were able to move first on vaccines and intellectual property,” he told The Africa Report. 

    The COVID-19 pandemic was proof of this. The countries that held early genomic sequence data acted faster, filed more patents, and kept more of the commercial benefits that came as a result. 

    Evaborhene also noted the national security dimension. “Access to pathogen and surveillance data is no longer just a public health issue, but is now tied directly to biosecurity, pharmaceutical innovation, and strategic advantage. It feeds into preparedness systems, military bio-defence and strategic planning,” he said. (The Africa Report)

    The U.S. also has historical precedent to draw on. PEPFAR, the President’s Emergency Plan for AIDS Relief, has operated across sub-Saharan Africa since 2003, funding health infrastructure including laboratories, electronic medical records, and disease surveillance networks.

    Those systems, which were built largely using American money, have generated huge datasets and there have been questions about who owns these data sets and benefits from them.

    The Clause That Changed Everything

    The most controversial part of the bilateral agreements is a data and pathogen-sharing obligation that somehow outlasts the actual MOUs.

    While the MOUs are designed to last two to five years, the commitments to share pathogen information and grant access to data systems extend well beyond that — reportedly 25 years for specimen and pathogen access, and 10 years of continued U.S. data access even after the agreement is terminated.

    Professor Oyewale, a former WHO regional virologist, made a very frank assessment. Writing in The Conversation Africa, he states that a unilateral termination clause which gives Washington the permission to pause or end programs that no longer benefit U.S. government interests reduces the sovereign partners to dependent states, contrary to claims of helping African nations become independent:

    “In particular, this US agreement has a unilateral termination clause that may leave African countries dangling on the tattered and shredded sovereignty of a beggar nation. On no account should Africa give raw data or pathogens for any amount of money or donation.”

    Ravi Ram from the People’s Health Movement, East and South Africa, pointed out the structural imbalance: “Bilateral agreements between the US and Kenya and other African countries are meant to limit the ability of any one partner country to challenge the terms of implementing the agreement, due to the unequal power relations.” (AllAfrica)

    Which Countries Are Pushing Back, and Why

    Zimbabwe, to the surprise of its citizens, rejected a deal worth $367 million over five years in late February 2026. The government cited data-sharing concerns and what it described as asymmetrical terms.

    The refusal was also connected to WHO Pathogen Access and Benefit-Sharing negotiations. There was an argument that signing bilaterally would unavoidably lead to African solidarity being undermined in multilateral forms.

    The consequences of this decision were immediate. 1.2 million men, women, and children receiving their HIV treatment and medication via U.S. donor programs were faced with a funding vacuum and there was some panic about medication shortages.

    Zambia dragged its feet after reports emerged (which Reuters said tied the deal to a ‘bilateral compact’ linked to mining collaboration proposed by Secretary Rubio), but the Zambian government stuck to its claims that there was no connection between the agreement and mineral deposits. Some civil society actors said that the data only flowed in one direction, from Zambia to the U.S. with no reciprocation.

    While Kenya’s Principal Secretary of Medical Services, Oluga Ouma, assured media the MOU contained no pathogen-sharing clauses, clause 3 of the document commits the country to exactly that. A court suspension was ordered, but it was still unclear if it would have a practical effect.

    Africa CDC’s Response

    The response from the Africa Centers for Disease Control and Prevention has been strategic. Director-General Jean Kaseya had launched the Africa Health Security and Sovereignty Agenda in November 2025, calling it “non-negotiable” (a new model in which African nations lead partnerships with clarity and confidence). Two months after that, 18 member states signed agreements that clearly contradicted that very agenda. (ICTworks)

    When Kaseya spoke at a virtual press conference in February 2026, he was quite direct about the stakes: “There are huge concerns regarding data, regarding pathogen sharing. We want to own our data in Africa. We want to own our future.” (Reuters via US News)

    But Kaseya refused an offer to serve as an observer during the bilateral negotiations, and his reason was respect for national sovereignty. He also still pledged technical support to every member state regardless of whatever choice they made. This is a position that clearly reflects the limits of an advisory body that operates without binding authority over 55 sovereign nations:

    “What we say to all African countries is negotiate based on your interest, policies and on the benefit that your country and the continent are getting.”

    The Scramble for Pathogens: A New Extractive Economy?

    Warwick University researchers Sharifah Sekalala, Shajoe J. Lake, Allan Maleche, and Timothy Wafula, writing in PLOS Global Public Health, have put forward strong arguments. They argue that what is happening is a structural reconfiguration of data ownership posing as partnership. They argue that this is all a farce. “Race is crucial to this arrangement,” they write. “Countries such as Kenya, Rwanda, and Uganda are asked to trade health data and pathogens for funding under the language of partnership, while the principal gains from expanded surveillance capacities, patent portfolios, and AI tools continue to accrue to the US and its allied markets.” (PLOS Global Public Health)

    They argue that the communities that generate the data are highly unlikely to hold intellectual property, shape research agendas, or reliably access the resulting products or rewards. This pattern is not new to Africa. Raw materials are extracted, and the value is added elsewhere.

    Citizens have been appealing to their governments and more than 50 civil society organizations published an open appeal urging African governments not to sign agreements that they described as designed to “dictate US terms that are not guided by African national interests”. (AllAfrica)

    What Happens When the Money Runs Out?

    The bilateral agreements also raise a structural problem that African health ministries are only beginning to grapple with. Some countries, such as Liberia, could face significant year-over-year drops in U.S. funding alongside sharp increases in co-financing expectations that exceed previous spending projections, according to analysis by the Council on Foreign Relations’ Think Global Health.

    In 2025, Nigeria’s Health Minister publicly confirmed that only about US$25,797 (₦36 million) of the ₦218 billion capital allocation for the Ministry of Health Headquarters was actually released for capital projects. The figure, confirmed by Minister Muhammad Ali Pate during his 2026 budget defence before the House of Representatives, refers to the Ministry’s Headquarters capital line — not the entire ₦2.38 trillion national health budget. Even so, it illustrates the chronic gap between what is allocated and what reaches the health system. Under the new bilateral framework, Nigeria must commit to a progressively increasing co-investment schedule from 2026 to 2030. The co-financing requirement is a legitimate long-term goal; in practice, for a country spending well under 6% of its budget on health, it is close to impossible to meet.

    But let’s not ignore that African governments have been more resilient than expected. Nigeria increased domestic health allocations by US$200 million within a month of Trump’s aid cuts; Ghana removed the cap on its National Health Insurance Levy to redirect full proceeds into the health sector. These moves signal a continent that is gradually taking ownership of its own future.

    So, Why Is the U.S. Really Interested?

    The official answer is pandemic preparedness, global health security, and the responsible transition away from aid dependency. None of that is false. Early access to pathogen data genuinely accelerates response time. Building African lab capacity genuinely serves global biosecurity. The U.S. is not simply manufacturing a pretext.

    But the unofficial answer, the one that civil society groups, virologists, and African health officials are now saying loudly, is that health data is a strategic asset, and the U.S. is using leverage it already holds (existing aid infrastructure, the threat of funding withdrawal) to lock in access before the rules of the game change at the WHO.

    The bilateral agreements are, among other things, a way to pre-empt the WHO Pathogen Access and Benefit-Sharing mechanism adopted in 2025’s Pandemic Agreement. They pose significant risks to the ongoing negotiations on that mechanism by replacing multilateral rules with bilateral arrangements the U.S. can terminate at will, in Professor Tomori’s assessment.

    The irony is sharp. A strategy framed around African self-reliance is, in its data clauses, engineering the opposite: a continent that generates the world’s most epidemiologically valuable data, but does not own it, cannot monetize it, and may not even be able to access the research it funds.

    “We want to own our data in Africa. We want to own our future. — Jean Kaseya, Director-General, Africa CDC”

    Sources & Further Reading

    • The Africa Report — Africa slams the door on extractive US health deals

    • The Conversation Africa — African countries are signing bilateral health deals with the US: virologist identifies the ‘red flags’

    • PLOS Global Public Health — America First, Africa Last? Health data deals and the new scramble for pathogens

    • KFF Tracker — America First MOU Bilateral Global Health Agreements

    • Think Global Health — Tracking the America First Bilateral Health Agreements

    • Reuters / US News — Africa CDC head cites major concerns over data, pathogen sharing in US health deals

    • ICTworks — African Digital Health Data Is a Condition of Global Health Funding

    • Health Policy Watch — December Deals: US Signs Bilateral Health Agreements With 14 African Countries

    • ODI — African leadership amid disruptions to US aid

    • Public Citizen — Letter to African Heads of State on Health Agreements (PDF)

    • AllAfrica — America First Global Health Strategy: A Framework for Co-Opting Healthcare in Africa?

  • SpaceX Is Going Public. Here’s Why It’s Already a Fight.

    SpaceX Is Going Public. Here’s Why It’s Already a Fight.

    Featured Image: SpaceX

    SpaceX is set to break records and make history later this month by taking the number one position as the biggest Initial Public Offering (IPO). SpaceX plans to sell 555.6 million shares at $135 apiece in a bid to break existing records by raising $75 billion. 

    The previous record was set at $35.4 billion  in 2019 by Saudi Aracmo which means that SpaceX is targeting to more than double that amount. 

    This move will get Elon Musk closer to being the world’s first trillionaire with the company aiming for an ambitious $1.75 trillion valuation.

    SpaceX is set to trade on the Nasdaq exchange and it’s speculated that shares may be listed as early as the 12th of June according to Reuters. 

    The primary coordinators are noted as Goldman Sachs, Bank of America Securities, Citigroup, Morgan Stanley, and JPMorgan.

    If you’re wondering, Elon Musk is not planning to sell any of his shares. After all this is done and dusted, his 42% equity stake will be worth roughly $735 billion, which puts him in the running to be the first trillionaire.

    A trillion is one million millions in case you were wondering. 

    Retail investors get a slice of the pie with a 30% allocation carved out through Fidelity, Robinhood, and Charles Schwab. This will be the biggest slice that goes directly to retail any IPO of this kind has ever offered. 

    Everyone else will buy on the open market day (June 12), and they will pay whatever price hype sets on the first day. 

    SpaceX Is More Than a Rocket Company 

    From another angle, the headline business is now the smaller piece of the story. SpaceX wrapped up a merger with Musk’s AI company in February of this year. This merger included Grok (the sometimes antisemitic chatbot), the AI infrastructure, and X which most of us still call Twitter.

    The plan to include X in the merger came across as unflattering. Ad revenue on the platform fell by roughly $100 million which looks quite bad if compared to other platforms like Meta and Reddit which experienced a growth in sales within the same period. 

    So in summary, Elon Musk bought Twitter in 2022 and changed its name, he then merged it with xAI, and now all this combined is now baked into the IPO prospectus.

    The revenue in 2025 came in at a total of $18.67 billion with a net loss of $4.9 billion. The loss can be traced back to the $3 billion that was sunk into the Starship R&D and xAI integration. 

    The company is taking a gamble by sinking money into infrastructure that isn’t currently in existence. 

    A valuation of $1.75 trillion will see investors paying about 100 times trailing revenue. Starlink makes sense and is profitable but everything else is just taking a gamble by betting on rockets, AI dominance, and the hope that space-based data centers will become a thing soon enough. 

    The Index Fund Problem 

    There’s a problem. SpaceX intends to float about 3-5% of its total shares at IPO. This means that by market cap, it would only rank behind Apple, Nvidia, Alphabet, Amazon, and Microsoft from the moment it lists.

    Trying to get a company this big into passive index funds normally takes quite a lot of time. The rules governing how this should happen were created in a world where IPOs were smaller and floats were larger. Companies also needed at least a year of public earnings before pension funds got involved. 

    SpaceX becoming part of the scene shook those rules and index providers have been trying to make adjustments since February when SpaceX decided to start pushing for faster inclusion. 

    Nasdaq was the first to make adjustments. As of the 1st of May, 2026, its new rules state that newly listed companies ranked in the top 40 by market cap are allowed to enter the Nasdaq-100 after only 15 trading days.

    The minimum float requirement was entirely removed for companies at that scale. SpaceX easily meets these requirements and will enter QQQ-tracking funds about three weeks after listing. 

    On the 26th of May, FTSE Russell followed by confirming a Fast Entry mechanism under which an IPO exceeding the Russell Top 500 market-cap breakpoint becomes eligible five trading days after listing, rather than waiting for the next quarterly review.

    SpaceX’s lockup schedule (with insider shares releasing in tranches tied to earnings reports, time milestones, and stock price triggers, and Musk himself locked up for 366 days) appears designed specifically to satisfy FTSE’s new float carve-out condition. Russell 1000 inclusion, previously expected in September, now happens roughly a week after the IPO.

    The S&P 500 is the only one still holding out. So far, it closed a consultation on May 28 proposing to cut the seasoning window from 12 months to six and eliminate the four-quarter profitability requirement for megacap companies.

    So far, no final decision has been announced. Under the accelerated proposal, a late-June IPO still wouldn’t make SpaceX eligible until around mid-December 2026. Full S&P 500 inclusion, the benchmark that most retirement money actually tracks, likely lands in late 2026 or early 2027, not days after listing.

    The Forced Buying Argument

    Bloomberg Intelligence analyst Rob Du Boff estimates that S&P 500 index funds could be required to acquire 19% of SpaceX’s available float within six months of inclusion. Russell 1000 and Nasdaq-100 tracked funds together would need to acquire 24% of available shares.

    This is mechanical rebalancing where every fund tracking those indexes has to buy SPCX on inclusion day, at whatever price the market has already pushed it to, because the rules say so.

    To accommodate the buying, funds will sell existing constituents. Apple, Microsoft, Nvidia — every company in the index gets slightly trimmed to make room. The scale of the forced rebalancing is what makes SpaceX’s IPO the first of its kind and unlike almost any that came before it. There’s a wall of passive capital that has to follow, regardless of valuation.

    The Pushback

    Not everyone is happy about any of this. Critics argue that bending index inclusion rules specifically for SpaceX sets a precedent with no clean stopping point. This is an issue because OpenAI and Anthropic are both reportedly considering 2026 IPOs at valuations that would trigger similar conversations.

    One market commentator quoted by Benzinga put it plainly: SpaceX hasn’t earned its seat at the table yet. The argument is that forcing passive fund managers to buy a low-float, unprofitable-at-the-aggregate-level company days after listing, at a 100x revenue multiple, because index providers rewrote their eligibility rules under pressure from bankers and asset managers, represents a structural distortion.

    Forced buying, as the argument goes, doesn’t mean SpaceX is worth $1.75 trillion. It means index funds have no choice.

    Morningstar made the case from the other direction. Any index claiming to represent the U.S. stock market that excludes one of the five largest companies by market cap is misrepresenting the market — the same argument was made, and eventually won, when Tesla spent months outside the S&P 500 while being enormous. The index, on that reading, has an obligation to hold SpaceX. The fight is over how fast.

    What Happens Next

    The roadshow runs through June 11 and pricing is set to happen that same evening. Trading opens June 12. Prediction markets are currently pricing a 94% probability of the offering closing inside the June window.

    Retail allocation is thin. If you’re planning to buy SPCX at the IPO price through Robinhood or Fidelity, you may get a small piece.

    Most retail participation happens in the open market, which means paying a premium on whatever momentum first-day buyers create.

    Analysts watching previous high-profile tech IPOs have noted that first-day pops frequently retrace 20–40% within 90 days. SpaceX’s first quarterly earnings as a public company are expected in early November.

    Musk holds 85.1% of the voting power through a dual-class share structure. Every institutional investor who buys SPCX — every index fund, every pension, every retail account — will own a piece of a company where no shareholder vote they cast has any practical bearing on anything. Board composition, executive pay, strategic direction: all of it sits with one person. That’s not unusual for founder-led tech companies at IPO. At $1.75 trillion, it’s just a larger version of something we have gotten used to.

    The record will fall in eight days. The argument about whether it should have been this easy to get there is already happening.

  • No Bribe, No Runaround: How Smart Police Stations Work

    No Bribe, No Runaround: How Smart Police Stations Work

    A group of men break into your home and brandish machetes in your face. They rob you blind. You walk to the police station, hands shaking, and you report the crime. There’s a short-lived pause before one of the police officers says, “Give us money for fuel and drinks before we can attend to you. If you can’t do that, go and bring the perpetrators yourself.”

    Now let’s switch scenes for a moment.  

    Marisela crossed the US-Mexico border on a tourist visa in 2019. By 2022 her status had lapsed, but she had been living in Houston, working, paying rent, keeping to herself. Then she was assaulted. She called a friend, not the police. She knew what calling the police meant for someone like her: a conversation about her documents before a conversation about what had happened to her. The man who assaulted her went on to assault someone else in the same building six weeks later. She found out later from a neighbour. She did not go back to the police.

    Her calculation was rational. A 2025 KFF/New York Times survey found that 41% of all immigrants, not just undocumented ones, personally worried that they or a family member could be detained or deported. When three quarters of a population are afraid a call for help might result in their deportation, the phone stays in the pocket. The perpetrator stays on the street.

    What Exactly Is a Smart Police Station?

    A smart police station is a digitally integrated law enforcement facility, fully or partially unmanned, where citizens interact with systems rather than front-desk officers. At its minimum, it is a self-service kiosk platform with guided digital workflows. At its maximum, it is a fully autonomous facility with AI-driven cameras, biometric identity verification, cloud-based case management, and real-time video links to remote command rooms.

    The concept merges e-government infrastructure, public safety technology, and (whether the architects admit it or not) corruption-by-design prevention. When there is no officer standing between a citizen and the system, there is no palmed-form moment or prejudice. The leverage point simply doesn’t exist.

    The shift in policing philosophy underneath all of this is from reactive to proactive. It’s from ‘respond after crime’ to ‘prevent, detect, and document in real time.’ That shift requires data infrastructure. And data infrastructure, it turns out, is also the best accountability system humans have yet devised.

    How It Actually Works: The Technical Architecture

    A smart police station is a layered stack of integrated systems. Here is what that stack looks like,  and what each component does in practice: 

    The Citizen Journey

    Step 1: Walk In (or Open the App).

    The citizen approaches a partitioned booth with a touch-enabled tablet screen, or opens a mobile app. There is no queue. There is no uniformed person to negotiate with.

    Step 2: Incident Classification.

    The interface presents categorised incident types: theft, assault, traffic incident, financial fraud, missing person, lost documents, and so on. CAD systems (Computer-Aided Dispatch) sit behind this layer. The moment an incident type is selected, the system begins creating a structured incident record linked to location, time, and category.

    Step 3: Identity Verification.

    Biometric scanners, typically FBI-certified fingerprint capture devices or passport/ID readers, verify the citizen’s identity without manual document handling. This creates an immutable link between the report and a verified identity, which is both a fraud deterrent and an evidentiary asset.

    Step 4: Evidence Submission.

    The system accepts structured digital input: typed statement, photo upload, video, audio recording, and GPS location tagging. In the EFPApp mobile version, live streaming to the remote command room is available, essentially a direct video link to an officer in real time. 

    Step 5: Remote Officer Review.

    Once submitted, a remote officer receives the report on a dashboard, flagged by incident type and AI priority scoring. For in-station kiosks, the officer appears on screen via encrypted video link, takes the statement verbally, asks follow-up questions, and if the situation warrants immediate response, dispatches patrol units. In Ethiopia’s Bole station, this entire interaction can happen in under ten minutes.

    Step 6: Case Record Creation.

    The report is automatically filed into the cloud-based Records Management System (RMS) with a timestamp, case number, and unique identifier that is visible to the submitting citizen. The record is immutable, it cannot be deleted by a single officer. It is cross-referenced with the national criminal database (NCIC/NLETS in the US; equivalent national registries in other jurisdictions). The fold-and-pocket manoeuvre is architecturally impossible.

    The Problem Smart Stations Are Solving (With Numbers)

    The Global Corruption Barometer (Africa) from Transparency International estimated that nearly 75 million people in Sub-Saharan Africa paid a bribe in a single year to access public services, and of the six public services surveyed, police contact had the highest bribery rate. An estimated one-in-ten bribery victims actually reported it. 

    Nigeria’s Global Corruption Barometer entry records that 44% of public service users paid a bribe in the previous twelve months. These statistics describe the modal experience of interacting with law enforcement across much of the continent.

    When reporting feels futile, victims don’t report. When crimes go unreported, perpetrators are not prosecuted. When perpetrators are not prosecuted, the incentive structure for crime remains intact. 

    Smart stations address the structural problem, not just its symptoms. Removing the human intermediary at the point of first contact removes the opportunity for dishonesty from the interaction entirely. 

    The Problem Is Not Only African

    The narrative around smart policing and digital crime reporting tends to get filed under ‘developing world governance reform’,  a frame that lets wealthier nations off the hook. The data does not support that framing. The US, the UK, and Western Europe have their own versions of the broken reporting loop, driven by different mechanisms but producing the same result: vast numbers of crimes that never enter the system, and perpetrators who stay free because victims had no realistic expectation of being helped.

    The Underreporting Problem in the United States

    According to RAINN, only 310 out of every 1,000 sexual assaults are reported to law enforcement. Nearly seven in ten survivors of sexual violence never file a report. The reasons are consistent across surveys: fear of retaliation, disbelief that police will act, shame, and fear of how the criminal justice system will treat them.

    Why Smart Stations Address This Too

    The architectural argument that applies to the first case applies equally to the second (Marisela), to the sexual assault survivor standing at a public counter, to any person whose decision about whether to report is shaped by what the officer behind the desk might do with their presence, their status, or their identity.

    A private booth with a timestamped digital submission does not ask about immigration status before opening a case. A remote officer receiving a report with the record already in the system has less opportunity and less incentive to let the conversation detour into territory unrelated to the crime. The immutable RMS record is as protective in Houston as it is in Addis Ababa: once it exists, it cannot be folded away. The accountability pressure runs both directions.

    This is not a claim that technology resolves structural racism, immigration enforcement policy, or decades of institutional distrust. It is a narrower, more defensible claim: removing the unmediated human interaction at the point of first contact removes one specific leverage point where vulnerable people are most exposed to the worst institutional outcomes. A system that logs every interaction, and makes that log visible to the submitting citizen, applies accountability pressure before the complaint reaches a human officer. It does not fix everything. It fixes something specific, something measurable, and something that affects millions of people in the wealthiest democracies on the planet, not just in Africa. 

    Ethiopia Steps Into History

    On February 9, 2026, Prime Minister Abiy Ahmed inaugurated Africa’s first fully unmanned smart police station in Addis Ababa’s Bole district, a commercial hub that functions roughly like a mid-tier African tech corridor. This is the fourth unmanned smart police station of its kind anywhere in the world.

    The project was developed in collaboration with the Ethiopian Artificial Intelligence Institute (EAII), which has been embedded in several of Ethiopia’s digital governance initiatives and is also responsible for the EFPApp (more on that shortly). The initiative sits within Ethiopia’s Digital Ethiopia 2030 strategy, a national blueprint targeting the digitisation of public institutions, improved service delivery, and technology-mediated access to government.

    According to Ethiopia’s state broadcaster ENA, the system runs 24/7/365 — no shift changes, no overnight staffing gaps, no ‘come back tomorrow.’ In its first week of operation, it received three reports: a lost passport and two financial fraud cases. Modest for a launch week. Proof of concept, technically speaking.

    The EFPApp: Smart Policing in Your Pocket

    Alongside the physical station, Ethiopia has been building the mobile layer. The EFPApp, developed by EAII for the Ethiopian Federal Police, is a citizen-facing crime reporting application available on both Google Play and the Apple App Store. Its feature set is worth reading in detail, because it illustrates what mobile-layer smart policing actually looks like at an engineering level:

    • Live streaming — stream video from your phone directly to the police in real time
    • Evidence upload — attach photos, videos, documents, or audio recordings to a report
    • Location sharing — GPS tagging with directions to the nearest police station or safe location
    • Report tracking — a case dashboard showing the status of submitted reports through investigation and prosecution
    • Free emergency calls — dial 991 directly from within the app at no charge

    In May 2025, the EFPApp won first place in the Best Police Mobile Application category at the World Police Summit in Dubai,  an international competition evaluated on innovation, design, technology, and real-world impact. The EFPApp currently serves Addis Ababa and Dire Dawa.

    Why This Technology Architecture Matters

    The Accountability Architecture of a Timestamped Record

    The most powerful thing a smart police station does is adding the timestamp. When a complaint enters a NIBRS-compliant RMS, that is, a records system conforming to the FBI’s National Incident-Based Reporting System standard, or its regional equivalent, it becomes an immutable entry in a shared database. Cross-agency queries are possible. The record can be retrieved by prosecutors. It has an author trail. A deletion or modification is itself logged.

    This is the structural intervention that bribe culture cannot survive. Corruption in police reporting happens in the gap between citizen and system, in the moment of verbal transaction. Close the gap with software and you close the opportunity.

    Privacy, Bias, and the Responsible Use Problem

    A fair technical write-up does not ignore the risks. AI use in law enforcement has a known and documented bias problem: facial recognition systems trained predominantly on lighter-skinned faces perform worse on darker-skinned individuals, a disparity that has led to wrongful identification events. Predictive policing models trained on historically biased arrest data risk encoding and amplifying existing discrimination. The US Department of Homeland Security’s 2025 AI Use Case Inventory reflects growing federal recognition that AI in public safety requires structured minimum-risk management practices, not just deployment guidance.

    In the African context, there is an additional concern specific to smart policing deployments: biometric data (fingerprints, facial scans, ID records) is being collected by government systems with, in many jurisdictions, limited independent oversight of how it is stored, shared, or used. The accountability infrastructure that prevents corruption at the front desk has to be matched by accountability infrastructure that prevents surveillance overreach at the back end. One without the other is just a different problem wearing better optics.

    A Station Where Your Voice Actually Counts

    What Ethiopia has built in Bole is a proof of concept that accountability can be designed into the architecture of a public institution, and it’s unfortunate that we haven’t been designing it that way all along. The corruption point does not merely become harder to act on, the system makes it structurally inaccessible. The timestamp is automatic. The case number is generated on submission. The remote officer is watching. The record cannot be quietly folded and pocketed.

    The question now is how to engineer for the inclusion gap: mobile-first interfaces with offline capability, kiosk UX designed for low-digital-literacy users, and biometric systems trained on the populations they serve.

    Sources & References

    1. Transparency International — Global Corruption Barometer Africa (9th Edition)

    2. Transparency International — Police corruption in Africa

    3. Transparency International — Corruption Perceptions Index 2025: Sub-Saharan Africa

    4. Governance journal (Wiley) — Police bribery and crime rates: Afrobarometer study

    5. ENA (Ethiopian state broadcaster) — PM Abiy launches smart police station

    6. ENA — Smart Police Station features and 24/7 operation

    7. ChimpReports — Ethiopia introduces digital police stations without officers

    8. Ethio Negari — Ethiopia launches first unmanned smart police station

    9. WutsHot — Ethiopia pilots smart police stations, Commander Demissie Yilma

    10. Open The Magazine — Ethiopia’s smart police station: connectivity challenges

    11. MEXC News / Technext — Ethiopian AI Institute collaboration detail

    12. AllAfrica — EFPApp selected Best Police Application, World Police Summit 2025

    13. Ethiopian Police University — EFPApp wins Best Police Mobile App, Dubai summit

    14. Google Play / Apple App Store — EFPApp feature listing

    15. Axon — Police RMS: complete guide

    16. Advanced Kiosks — Smart Police Station platform (Decatur PD, Itasca County SO)

    17. PSPortals — Police department software guide 2026 (Mark43, CivicRMS, PremierOne)

    18. GlobeNewswire / Research and Markets — Law enforcement software market: $33B by 2030

    19. Veritone — Modern police technology tools and trends

    20. Critical TS — New police technology 2026 (AI video analytics, LTE/5G, FirstNet)

    21. Police Records Management (PRI) — State of AI in law enforcement records 2025

    22. DHS — 2025 AI Use Case Inventory (minimum risk management practices)

    23. Fulcrum Biometrics — Biometrics in law enforcement: LiveScan and ID verification

  • How Pope Leo XIV Is Taking On Big Tech

    How Pope Leo XIV Is Taking On Big Tech

    Something unusual happened at the Vatican on Monday.

    Pope Leo XIV walked into the Synod Hall not to address cardinals or lead a prayer as most might assume, but to stand beside a Silicon Valley co-founder and launch the most ambitious papal document in a generation. 

    Featured image: Matthew Schwartz (UnSplash)

    A Pope, a Tech Co-Founder, and a Vatican Auditorium

    Pope Leo XIV stepped into the main auditorium of the Synod Hall to launch his first encyclical, a formal papal teaching document, standing beside Chris Olah, co-founder of Anthropic, the American AI company behind Claude. The pairing was deliberate, symbolic, and quite extraordinary by Vatican standards.

    The encyclical, Magnifica Humanitas (“Magnificent Humanity”), runs to 245 paragraphs and centers on what Pope Leo XIV considers the defining moral question of the age: how humanity governs artificial intelligence before it governs us. He signed the document on May 15, 135 years to the day after his 19th-century namesake, Pope Leo XIII, signed Rerum Novarum, the landmark text that established the Church’s teachings on workers’ rights during the Industrial Revolution. That date was chosen deliberately.

    One Vatican official privately described the decision to invite Olah as unusual and said it should be read as a signal of Pope Leo’s seriousness. “We haven’t usually invited someone from the outside,” the source told the National Catholic Reporter. Another Vatican statement made the boundaries clear: Anthropic’s inclusion was “not an endorsement, prize, reward or canonization.” It was, rather, an open invitation to dialogue.

    Why Anthropic? Why Now?

    The choice of Anthropic over OpenAI or Google carried layers of meaning. As CNN reported, it also reflects the ongoing friction between Pope Leo XIV and the Trump administration, which has taken a hands-off approach to AI regulation. Anthropic has been in a legal dispute with Washington over the use of its technology in military and surveillance applications, a tension that sits squarely within the encyclical’s warnings about autonomous weapons.

    Anthropic has cultivated ties with the Vatican for some time. The National Catholic Reporter noted that the company listed three Catholic thinkers, among them Bishop Paul Tighe, Secretary of the Dicastery for Culture and Education, among contributors to its Claude Constitution, the internal ethical framework that guides how its flagship AI model behaves.

    For his part, Olah used his platform at the Vatican to say something striking for a Silicon Valley figure: that people outside the incentive structures of AI labs are essential. “Every frontier AI lab, including Anthropic, operates inside a set of incentives and constraints that can sometimes conflict with doing the right thing,” he told the audience. “That is why, if we want this technology to go well, it is enormously important that there be people outside those incentives.” He described the encyclical as exactly that kind of outside voice.

    The Math-Trained Pope Who Saw This Coming

    Pope Leo XIV, born Robert Francis Prevost in Chicago in 1955,  is the first American pope in history, elected on May 8, 2025. Before entering the priesthood and the Augustinian order, he studied mathematics: a background that may explain why his engagement with AI has felt more grounded than that of previous pontiffs.

    From his very first formal address to the College of Cardinals, Pope Leo XIV identified artificial intelligence as one of the defining challenges of his papacy, naming it a threat to “human dignity, justice, and labour.” He even cited AI as one reason for choosing the name Leo, explicitly drawing the parallel between industrialisation in the 1890s and the technological transformation underway today.

    At the Second Annual Rome Conference on AI, Ethics, and Corporate Governance in June 2025, he warned that AI “must never forget human dignity” and cautioned about the technology’s effects on children and young people. Days before releasing the encyclical, he established a dedicated Vatican commission on AI. 

    His predecessor, Pope Francis, laid some of this groundwork. In January 2025, the Vatican released Antiqua et Nova, a lengthy theological note on AI’s promises and risks, signed off on by Francis before his death in April 2025. The Church has, in fact, been monitoring AI’s development for over four decades through the Pontifical Academy of Sciences. But Magnifica Humanitas goes further than anything that came before it.

    What the Document Says: The Case For AI

    Pope Leo XIV came to this document as an enthusiast as well as a critic, and the encyclical gives real weight to what AI can offer. The document describes genuine enthusiasm for what the technology could do in healthcare, environmental monitoring, scientific research, and development, areas where precision and scale could save lives. It encourages progress and frames AI, when used well, as “part of the collaboration of man and woman with God.”

    The encyclical acknowledges the “immense potential” of the technology. AI applications in medicine get particular attention, consistent with Pope Leo’s message to the Pontifical Academy for Life, where he spoke warmly about diagnostic tools that can “alleviate immense suffering.” The encyclical does not condemn AI; it demands that humanity govern it.

    What the Document Says: The Case Against

    The warnings in Magnifica Humanitas are sharper and more detailed than anything a sitting pope has put in writing on this subject. Several themes dominate.

    On war: Pope Leo’s most direct condemnation is reserved for autonomous weapons. Drawing on a line that will likely be quoted for years, he wrote: “No algorithm can make war morally acceptable.” AI, he argues, accelerates the horror of conflict and strips it of human accountability. He calls for the technology to be removed entirely from lethal decision-making chains.

    On inequality: “AI tends to amplify the power of those who already possess economic resources, expertise and access to data,” he wrote. The encyclical warns that a handful of wealthy individuals should not be allowed to determine humanity’s future, and flags the growing gap between those who can participate in the digital revolution and those left entirely behind.

    On power concentration: Pope Leo XIV takes direct aim at Big Tech, calling for AI to be freed from “monopolistic control” and demanding stronger state and international regulation. The encyclical calls for government oversight of AI companies, a direct rebuke of the current laissez-faire approach in Washington.

    On the environment: The document flags AI’s overwhelming resource demands that compete with human needs. Training large models and running the hardware required to support them consumes vast amounts of energy and water, significantly contributing to CO2 emissions and straining infrastructure, particularly in developing regions.

    On human identity: Perhaps most philosophically, Pope Leo XIV pushes back against the very framing of “artificial intelligence” as a term. Quoting Pope Francis, the document argues that the word intelligence as applied to machines “can prove misleading”, and AI is something produced by human intelligence, not a form of it. The soul, conscience, and moral judgement that define the human person cannot be replicated or outsourced.

    “Disarm” — The Word at the Heart of It All

    The encyclical is built around a single, unusual word: disarm. In paragraph 110, Pope Leo XIV writes that AI is no longer just a tool but “an environment in which we are immersed and a power with which we must contend” . He also emphasized that mere regulation falls short.

    “Disarming AI means freeing it from the mentality of ‘armed’ competition, which today is not limited simply to the military context, but is also an economic and cognitive phenomenon,” he wrote. The word echoes throughout the document as a kind of refrain: disarming the race for algorithms, disarming economic competition, disarming the logic of domination.

    But disarmament, in Pope Leo’s framing, is only the beginning. “Disarming is not enough, we must build,” he adds, calling for broad, inclusive participation in how AI is programmed, governed, and distributed.

    A Bigger Stage Than Expected

    The presentation itself broke with Vatican tradition. Encyclicals are normally handed off to cardinals to introduce in a press room; Pope Leo XIV oversaw the release himself, addressing the world from the main Vatican auditorium alongside a lineup that included Cardinals Víctor Manuel Fernández and Michael Czerny, theologians Anna Rowlands and Leocadie Lushombo, and Olah.

    The drafting process, Pope Leo XIV explained, began in July 2025 at Castel Gandolfo. “I have listened to scientists and engineers who work with sincere enthusiasm on technologies capable of alleviating immense suffering, to political leaders and public officials who have perseveringly sought just rules, to parents and teachers who are deeply concerned for the future of younger generations,” he said at the launch.

    For those watching from the tech industry, the encyclical lands at a complicated moment. The Trump administration has shelved meaningful federal AI oversight, Anthropic is fighting in court over how its models are used by the military, and AI governance globally remains patchy and contested. In that vacuum, a 245-paragraph document from a pope with a mathematics background, standing next to one of Silicon Valley’s most safety-focused voices so far, is at the very least an unusual intervention.

    Whether or not one shares Pope Leo’s faith, the question of who regulates the people building the most powerful technology in human history is one that reaches well beyond the Vatican walls.

    Sources

    National Catholic Reporter – Pope Leo, Anthropic co-founder call for church-tech ethics partnership

    Anthropic – Chris Olah’s remarks at the encyclical presentation

    NPR – Pope Leo takes aim at big tech in sweeping encyclical on AI

    CNN – Pope Leo warns of AI fueling warfare in first major theological document

    PBS NewsHour – Pope Leo XIV to launch encyclical on AI with Anthropic’s co-founder

    Religion News Service – Pope Leo says AI must serve humanity, not the powerful few

    Vatican News – Magnifica Humanitas announcement

    Fortune – Pope Leo launches an AI commission

    Catholic Digest – Pope Leo XIV’s appeal to ‘disarm AI’

    Encyclopaedia Britannica – Leo XIV

    USCCB – Morality of AI depends on human choices

    Catholic World Report – Pope Leo XIV unveils his encyclical

  • The AI Bill Has Come Due: How the Cost Crisis Is Reshaping the Tech Industry

    The AI Bill Has Come Due: How the Cost Crisis Is Reshaping the Tech Industry

    For the better part of two years, the story in tech was simple: AI makes your engineers faster, happier, and more productive. Companies handed out access to AI coding tools the way offices once handed out free snacks. Generous, enthusiastic, and completely without a plan for what the bill would look like. That bill has now arrived, and nobody is particularly thrilled about the number on it.

    Featured Image: Igor Omilaev (UnSplash)

    What started as a few raised eyebrows at expense reports has turned into a full-blown industry reckoning. Microsoft, Uber, and GitHub are the names making headlines right now, but they represent a much wider pattern: the gap between what companies expected AI to cost and what it actually costs at scale is proving to be wider than anticipated.

    Microsoft Pulls the Plug on Its Own Experiment

    In December 2025, Microsoft gave thousands of its internal employees access to Claude Code, Anthropic’s AI-powered coding tool. The rollout covered the Experiences and Devices division, the team responsible for building Windows, Microsoft 365, Outlook, Teams, and Surface. Access went beyond developers to include designers, product managers, and non-technical staff, as part of a push to let people without coding backgrounds build tools using AI.

    It worked, perhaps too well. According to The Verge’s Tom Warren, Claude Code became “very popular, perhaps a little too popular” inside Microsoft over the following six months. Engineers preferred it over GitHub Copilot, Microsoft’s own product. Which is, to put it diplomatically, an awkward outcome when you are Microsoft.

    By May 14, 2026, Microsoft began cancelling those Claude Code licenses, with a full transition deadline of June 30 that conveniently lines up with the end of Microsoft’s fiscal year. Engineers are now being directed to GitHub Copilot CLI instead. Rajesh Jha, Executive Vice President of Microsoft’s Experiences and Devices group, framed the decision in an internal memo: “When we began offering both Copilot CLI and Claude Code, our goal was to learn quickly, benchmark the tools in real engineering workflows, and understand what best supported our teams. Claude Code was an important part of that learning… Copilot CLI has given us something especially important: a product we can help shape directly with GitHub.”

    The decision is partly about cost, partly about the optics of your own workforce preferring a rival tool, and partly about platform consolidation. On the Q3 earnings call, CEO Satya Nadella confirmed the broader shift: “Any per-user business of ours, whether it’s productivity, coding, security, will become a per-user and usage business.” The direction of travel is clear.

    It’s worth emphasizing that this doesn’t mean Microsoft has walked away from Anthropic. The two companies have a substantial partnership in place, with Microsoft committing to invest up to $5 billion in Anthropic and Anthropic agreeing to purchase $30 billion in Azure compute capacity. Claude models remain available through Microsoft Foundry and within Microsoft 365 Copilot for specific tasks. The license cancellation is exactly what it looks like, a cost management move.

    Uber Spent Its Entire Year’s AI Budget in Four Months

    Uber’s situation is more straightforward, and more alarming. The ride-hailing company also rolled out Claude Code to its engineering team in December 2025. By March 2026, 84 percent of Uber’s roughly 5,000 engineers were classified as active agentic coding users. By April, roughly 95 percent of engineers were using AI tools every month, and around 70 percent of all committed code was coming from those tools.

    The problem is that agentic AI tools don’t work like traditional software subscriptions. When an engineer uses Claude Code for a multi-step autonomous coding session, the tool reads through entire codebases, writes tests, runs them, catches failures, and iterates through multiple passes. Each of those steps consumes tokens. Token costs add up. Monthly API costs per engineer at Uber ranged from $500 to $2,000, depending on how heavily they used the tool.

    Uber CTO Praveen Neppalli Naga told The Information that the company had burned through its entire planned 2026 AI budget by April. Four months into the year. “I’m back to the drawing board,” Naga said, “because the budget I thought I would need is blown away already.” The finance team’s original model had been built around fixed seat counts and low-frequency usage. It completely failed to anticipate what happens when thousands of engineers adopt an agentic tool as their primary workflow.

    Uber’s COO Andrew Macdonald later described a “head-exploding moment” when Naga’s comments went viral, sparking internal conversations about whether higher token usage was actually translating into a proportional increase in useful consumer features. For now, Macdonald said it was becoming harder to justify the costs. Uber is exploring OpenAI’s Codex alongside Claude Code as it searches for a cost structure that works at its scale. The internal leaderboards that once ranked engineers by AI tool usage in a trend known as “tokenmaxxing”, are likely being viewed through a slightly different lens.

    GitHub Rewrites the Pricing Contract

    GitHub, which is owned by Microsoft, announced in April 2026 that all Copilot plans will move to usage-based billing starting June 1, 2026. Under the old model, a quick chat question and a multi-hour autonomous coding session cost the user the same amount. That pricing structure made sense when Copilot was primarily an inline code suggestion tool. It stopped making sense when Copilot became an agentic platform running long, multi-step coding sessions across entire repositories.

    GitHub’s Chief Product Officer Mario Rodriguez didn’t dress it up: “GitHub has absorbed much of the escalating inference cost behind that usage, but the current premium request model is no longer sustainable.” The new system replaces premium request units with GitHub AI Credits, where one credit equals one cent and usage is calculated based on token consumption. Existing monthly plans migrate automatically; annual subscribers will face price increases as their plans expire.

    The broader message is that the AI tool industry is moving toward a model where your bill reflects what you actually used, not what you paid for up front. For individual developers, that means more transparency. For enterprise IT and finance departments managing thousands of seats, it means unpredictable monthly invoices and a new cost category they’ll need to track, forecast, and explain to the board.

    The Wider Picture: An Industry-Wide Reckoning

    Microsoft and Uber are the loudest examples right now, but they’re far from alone. AI software prices across the industry have reportedly climbed between 20 and 37 percent as the real economics of running agentic tools at scale become clearer. The original pitch for AI coding tools was that they’d pay for themselves through productivity gains. That math is being stress-tested by actual usage data.

    The underlying problem is structural. Traditional enterprise software budgets are built around seats and licences: you pay a fixed amount per user per month, and finance can model that out without much drama. Token-based billing introduces a variable that existing procurement and FinOps teams just aren’t set up to manage. Analysis from EPC Group found that 81 percent of enterprise leaders are worried about AI vendor dependency, yet only 6 percent say they could switch providers without significant disruption. That combination of dependence and unpreparedness is not a comfortable position to be in.

    For smaller companies and startups, the implications are more immediate. If a company the size of Microsoft can have its engineering budget blindsided by AI usage costs, a startup with fifteen engineers and no dedicated FinOps function faces the same problem at a scale that could genuinely be existential. The tools are powerful. They are also, at high adoption rates, expensive in ways that flat-rate pricing never revealed.

    What This Means for Employment

    The employment picture here is more complicated than it first appears, and it cuts in two directions.

    One argument says AI is proving too expensive to actually replace human workers at scale. If running AI agents costs $500 to $2,000 per engineer per month, on top of that engineer’s salary, the tool is an amplifier, not a substitute. At some organizations, compute costs now exceed the salaries of the human teams the tools are supposedly replacing. This is AI making skilled workers more central, not less so, because someone still has to govern, direct, and clean up after the agents.

    The counterargument is that cost pressure tends to find its outlet somewhere, and labor is often where it lands. Marell Evans, founder of Exceptional Capital, predicted in late 2025 that as AI budgets grow, human labor budgets will be cut to compensate. Rajeev Dham of Sapphire Ventures agreed that 2026 would mark real shifts in how companies allocate resources between people and tools. Jason Mendel of Battery Ventures described 2026 as the year AI would move from augmenting workers to automating work itself in certain categories.

    What’s happening right now looks less like replacement and more like compression. Fewer engineers may be needed to maintain or ship the same volume of work as AI tooling matures. Entry-level and mid-level roles that once involved writing significant amounts of routine code are already seeing fewer postings. The demand for engineers who can direct, review, and architect AI-generated output is growing. The demand for engineers whose main job is writing that output is contracting.

    There’s also the question of what Microsoft’s internal mandate means in practice. When a company tells its engineers to stop using the tool they prefer and switch to a different one, productivity takes a short-term hit. Engineers who built workflows around Claude Code now need to rebuild them around Copilot CLI. That kind of transition is rarely frictionless, and the people closest to it tend to have opinions about it that don’t always make it into the press release.

    What the Numbers Actually Show

    None of this means AI is failing. The Uber story, read carefully, is a story about a tool that worked so well engineers couldn’t stop using it. The problem was that nobody modeled what widespread adoption of a genuinely useful product would actually cost.

    The AI industry is going through a maturation that every major technology has gone through: the gap between early adopter enthusiasm and sustainable business deployment. Cloud computing went through it. Mobile went through it. AI is going through it now, just faster and louder because the sums of money involved are large and the claims made on the way up were ambitious.

    Companies aren’t abandoning AI. They’re getting more serious about it. There’s a difference between those two things, even if the budget memos look similar from the outside. The next phase of enterprise AI will look less like unlimited access for everyone and more like deliberate, governed deployment with actual cost controls. That’s probably a more durable model, even if it’s a harder one to sell internally.

    For engineers watching all of this, the clearest takeaway is that the tools aren’t going away, but neither is the expectation that you know how to use them well. The premium is shifting toward people who can evaluate AI output critically and understand where it adds genuine value, rather than just reach for it by default. That’s a different skill than writing fast code, and it’s the one most relevant right now.