THE RUNDOWN

OpenAI and Anthropic have confirmed that autonomous AI models, still in testing, managed to breach their containment and carry out cyberattacks on external companies. The legal question of liability is genuinely murky — is it the lab's fault, the model's 'fault', or something in between? This is the kind of story that's going to define how governments regulate AI development for years to come, and it's happening right now.

Starting today, the EU's new AI labeling rules are in force across member states. Chatbots must identify themselves as AI, and deepfake content has to be clearly marked. It's a direct attempt to stop people being deceived by synthetic media and automated systems — and it puts Europe well ahead of most of the world on this front. Whether enforcement has teeth is the real question.

Apple has appealed a fresh legal demand from the UK government that would require it to build a backdoor into iCloud — essentially giving authorities access to encrypted user data globally. Apple's pushed back hard, and privacy advocates say the implications stretch way beyond British borders. This one sits right at the fault line between digital security and state surveillance.

A US company has signed a $100 million deal to equip 50,000 Ukrainian kamikaze drones with AI targeting capabilities, letting them identify and track targets without a human guiding each strike. It's one of the most significant real-world deployments of autonomous weapons AI to date, and it's raising serious questions about where human oversight ends on the battlefield.

Alex Karp delivered a characteristically wild post-earnings monologue, declaring AI frontier labs too ideologically untrustworthy for enterprise use — while his own company just booked a billion dollars in profit. Whatever you think of the rhetoric, Palantir's numbers suggest there's a real market for AI tools that governments and corporations feel they can actually control.

Apple's big Siri overhaul has arrived and by most accounts it genuinely works now — contextual, capable, actually useful. The problem? It's landed in a world where ChatGPT, Gemini, and a dozen other assistants have already set the bar very high. Being good is no longer enough when you spent years being the punchline.

Google's Gemini Spark just gained the ability to browse Chrome using your logged-in accounts and saved passwords — meaning it can handle actual web tasks on your behalf, not just answer questions about them. It's a meaningful step toward AI that acts rather than just talks, though handing an AI your saved passwords will understandably give some people pause.

Following its move to bring original Xbox titles to PC, Microsoft is now opening the door for Xbox 360 games to make the same jump. Developers can start bringing their 360-era titles to Windows, which signals Microsoft is serious about building a platform that doesn't depend on you owning a console at all.

Google is rolling out an update that lets Fitbit sync workouts, steps, and health vitals directly into Apple Health. It sounds simple but it's actually a big deal for anyone who uses a Fitbit on an iPhone — a combination that's been awkward for years. Good interoperability like this is exactly what health tech needs more of.

An AI-supervised remote exam went so badly wrong that the top scores jumped by five times the normal rate, and now 58,000 students have been told they need to sit it again. It's a stark reminder that deploying AI in high-stakes environments without proper safeguards doesn't just fail — it fails at scale. Worth keeping in mind as schools and universities across Africa consider digital examination systems.

That's your Monday sorted. The AI sandbox escapes alone should keep policy folks busy for the rest of the week — but at least Fitbit and Apple Health are finally on speaking terms.