THE RUNDOWN

OpenAI says it's pausing development work on an internal AI model called Astra after it crossed what the company calls a 'critical cybersecurity threshold' — meaning it could independently identify and carry out cyberattacks on its own. That's not a feature; that's a threat. The decision signals that even OpenAI is starting to take its own security red lines seriously, though it raises obvious questions about what else is in the lab that hasn't tripped any alarms yet.

If you own a Framework laptop, your name, email, phone number, and home address are likely in the hands of hackers. The modular computer maker confirmed this week that its entire customer database was accessed in a breach — though it says no payment information was exposed. It's cold comfort for customers who trusted a privacy-forward brand, and a reminder that even companies doing everything right on hardware can still get burned on data security.

A New Mexico judge ruled this week that Meta created a 'public nuisance' by exposing children to harmful content on its platforms, ordering the company to fund a $567 million mental health treatment programme. It's one of the most significant legal blows Meta has taken in the ongoing global conversation about social media and young people. Expect more states — and more countries — to be watching this ruling closely.

The Trump administration has now paid out nearly $4 billion in cancellation fees to convince developers to abandon 12 offshore wind projects, with the latest deal alone costing taxpayers $1.2 billion. That's a staggering sum to spend on not building clean energy infrastructure. Critics are pointing out that this is money leaving government coffers to actively shrink the renewable energy sector.

Cloudflare launched Kitesurf this week — a cloud-hosted browser that's designed not for humans but for AI agents doing automated tasks. It's lighter than Chromium and purpose-built for the kind of web scraping, form-filling, and navigation that AI systems increasingly need to do on their own. It's a quiet but significant infrastructure move that tells you a lot about where the web is heading.

Disney is rolling out a natural language search tool for Disney Plus that lets you describe what you're in the mood for — by typing, speaking, or picking a suggested prompt — and get a personalised row of content back. It sounds small, but it's a direct shot at the endless scroll problem that makes streaming exhausting. Whether it actually works better than a keyword search is the real test.

Microsoft Edge is dropping support for the Manifest V2 extensions platform, which means popular ad blockers like uBlock Origin will stop working — exactly what happened when Chrome made the same move. If you rely on an ad blocker in Edge, you'll need to find a Manifest V3-compatible alternative, and fast. The pattern here is clear: the big browser makers are quietly squeezing out the tools that users actually want.

Security researchers who scanned the Polish web found serious vulnerabilities in critical public infrastructure — including hospitals, courts, and airports — with common content management software serving as the main weak point. The findings are a sharp reminder that government IT isn't immune to the kind of lazy patch management that plagues the private sector, and that cybersecurity gaps in public services carry very real human costs.

A broader picture emerging from the security world this week: hackers are now using AI tools to find and exploit vulnerabilities that didn't meaningfully exist before — including gaps that defenders haven't even thought to look for yet. The speed advantage is shifting, and traditional reactive security approaches are struggling to keep up with automated threat generation.

Europe's free satellite imagery service, Copernicus, just added wildfire visualisation to its browser — letting anyone track active fires and burned zones as they happen. With this summer already breaking wildfire records across multiple continents, it's the kind of open-access tool that emergency services, journalists, and researchers across Africa and beyond should have bookmarked right now.

That's your Friday wrapped. From OpenAI hitting its own emergency brake to Meta writing a nine-figure cheque for the harm its platform caused kids — it's been a week that makes you think hard about who's actually in control of the technology shaping our lives.