THE RUNDOWN

The preparedness team wasn't decorative — it was supposed to be the internal check on whether OpenAI's models were becoming genuinely dangerous before they shipped. Disbanding it quietly, with no public statement, raises real questions about who exactly is asking the hard questions inside that organisation now. For anyone following developments in AI governance, this is a significant moment.

Meanwhile, ChatGPT's macOS desktop app has rolled out a feature called Computer History, which tracks your clicks, keystrokes and on-screen activity to learn your working habits and suggest automations. OpenAI frames it as a productivity tool, but what it actually means is that your private work sessions are becoming training data. You'll want to check your settings before that feature quietly switches on.

A woman has come forward claiming her stepfather used xAI's Grok chatbot to transform a childhood photo of her into explicit imagery. She described AI tools as 'taking everyday life and turning it into child sexual abuse.' It's a deeply disturbing case that underscores exactly why AI safety guardrails — the kind OpenAI just dismantled — actually matter in practice, not just on paper.

Dario Amodei pushed back this week against critics who say he's been painting AI's future too darkly, arguing that the public backlash against AI is 'fundamentally a crisis of trust.' He's not wrong that trust is at the centre of it — but disbanding safety teams and building screen-recording features into chatbots doesn't exactly help rebuild it. The security and privacy concerns around AI are now mainstream, not fringe.

Here's the money story of the day: Stripe is reportedly in talks to acquire OpenRouter, an AI gateway startup, for more than $7 billion. OpenRouter's own CEO recently described the company as 'Stripe for AI' — essentially a unified routing layer that lets developers tap into multiple AI models through a single API. If this deal closes, it positions Stripe as a major piece of plumbing for how AI gets built and paid for at scale.

Mark Zuckerberg has been selling his version of an AI-first future hard — but according to TechCrunch's Equity podcast, a lot of people simply aren't buying it. There's a growing sense that Meta's AI push is more about catching up to competitors than offering a coherent vision that genuinely improves people's lives. The gap between the pitch and the product is starting to show.

After nearly a decade of iteration, reviewers are calling the Samsung Galaxy Z Fold8 and Fold8 Ultra genuinely good phones — not just impressive engineering experiments. The verdict is that the form factor has finally matured into something that works as a daily driver, not just a conversation starter. If you've been watching foldables from the sidelines, this might be the generation worth paying attention to over on our phones coverage.

Google has built some genuinely smart theft-detection features into Android over the past couple of years — but most people have no idea they exist, let alone that they're switched off by default. Features like Theft Detection Lock use your phone's motion sensors to detect if it's been snatched and lock the screen automatically. Worth two minutes of your time to enable them.

On a completely different note: the James Webb Space Telescope has helped astronomers identify what may be a 'black hole star' — an object that lends credence to a long-theorised explanation for mysterious red spots observed in the early universe. These hypothetical stars are thought to be powered by a black hole at their core rather than nuclear fusion, and finding evidence for one is a genuinely big deal in astrophysics.

The dream of hopping into an electric air taxi for your morning commute is shifting again — TechCrunch's Mobility team reports that the path to commercialisation for eVTOL aircraft is looking bumpier than the industry projected even two years ago. Regulatory hurdles, battery limitations and the sheer cost of certification are all pushing timelines further out. The future of urban air travel is still coming, just not as fast as the brochures suggested.

It's been one of those days where the AI safety stories hit harder because they're not hypothetical anymore — they're showing up in court cases, in deleted teams, and in features that quietly record your screen. Stay sharp out there.