This week, AI’s biggest fight didn’t happen in a lab demo or a Senate hearing — it happened in eight separate state courtrooms, where attorneys general decided ChatGPT’s safety promises deserved a judge’s opinion rather than just a press release.
It made for a fitting backdrop to a week that was long on real, checkable progress and short on patience for vibes-based panic. A scam-detection model that never leaves your phone quietly shipped in Singapore, Meta and Apple solved the unglamorous problem of running a genuinely huge model on hardware you can actually own, and a cybersecurity startup got bought specifically because it’s good at finding bugs before attackers do. Meanwhile, a Forbes contributor announced that superintelligence had already arrived and nobody noticed — which, to be fair, would explain a lot of comment sections.
Worth Reading
- Eight states sue OpenAI over ChatGPT safety concerns — Iowa, Arizona, and six other states filed separate lawsuits this week alleging OpenAI marketed ChatGPT as safe for kids while it exposed vulnerable users to dangerous advice, which is proof that AI oversight isn’t just happening in hearings anymore — it’s happening in courtrooms with real penalties attached.
- Google and Singapore Develop AI Scam Detector That Works Entirely on Your Phone — GovTech Singapore and Google DeepMind fine-tuned a lightweight Gemma model that flags manipulative scam conversations in real time without ever sending your private texts to a server, which is the rare AI-safety story that’s actually shipping rather than just promising.
- Introducing Muse Spark Open — Meta released an 816-billion-parameter open-weight model and worked with Apple to get it running locally across a pair of Mac Studios instead of a cloud GPU farm, a genuinely useful step toward powerful AI that lives on your desk instead of someone else’s server.
- Cohere announces plan to buy European AI cybersecurity startup Aisle — Cohere is acquiring a startup whose AI has already uncovered more than 500 previously unknown software vulnerabilities, folding bug-hunting AI directly into a frontier model company’s toolkit instead of treating security as an afterthought.
- Nvidia CEO Jensen Huang said he would cut off chip supply to unsafe AI developers — Huang paired that hypothetical threat with a genuine $250 million pledge to UK AI-safety research, a useful reminder that taking safety seriously and refusing to catastrophize about it are not actually opposites.
Spot the Hype
- Will A.I. end humanity? — A local news segment polled viewers’ fears, then asked safety researchers to confirm them, and landed bravely on “maybe the robots will get us” — truly fearless journalism.
- Superintelligence Is Already Here—We’re Too Distracted By AI To See It — A contributor declares superintelligence has quietly arrived and slipped past all of us, which would also explain why it still can’t reliably summarize a long email.
- “Very few businesses are planning for superintelligence”: Preparing for post-human work — This one frets that offices aren’t ready for a machine-run “post-human workplace,” a scenario currently backed by roughly the same amount of evidence as your coworker’s lottery strategy.
What’s the most breathless AI headline you ran into this week — and which of today’s actual stories has you genuinely interested? Tell us in the comments, and if you missed it, catch last digest here: A Mind-Controlled Arm Felt a Strawberry. The Internet Said Your Fridge Might Rebel.
Leave a Reply