The past month in AI has been unusually calm, at least by this industry's standards — labs prioritizing safety reviews over splashy releases after a string of security scares. That calm broke today.
Anthropic released Claude Fable 5.1, an upgrade squarely aimed at the complaints that piled up around the original Fable 5. The new model now tops Artificial Analysis's Intelligence Index with a record score of 66, more than doubling its predecessor's results on scientific research benchmarks and pulling ahead on knowledge work generally. Anthropic says typical tasks now cost roughly 25% less to run — though independent testing found that max-effort tasks actually run about 20% more expensive, since the model writes nearly twice as much text per answer. The more consequential fix is behavioral: Fable 5's safety filter, a major source of user frustration, now intervenes 60% less often on cybersecurity work and a striking 85% less on routine medical and biology questions. Anthropic also quietly released a lower-guardrail sibling, Mythos 5.1, though it's currently restricted to vetted U.S. cybersecurity and biology researchers. The timing isn't a coincidence — OpenAI reportedly has its own next-generation model, Astra, coming as soon as this week, after restarting a training run it had frozen following a security breach earlier this summer. Astra is being classified as OpenAI's first model with a "Critical" cyber risk rating, which tells you something about how seriously these labs are now treating what they're about to ship.
Politics found its way into the AI conversation today too, in an unusual place. Senator Bernie Sanders published an op-ed in Fox News calling for a global pause on frontier AI development, arguing that even the executives building these systems admit the technology is "escaping their control." His list of concerns spans energy consumption, environmental damage, and children's mental health, along with a prediction that AI could eliminate tens of millions of jobs within a decade. He's citing the past month's parade of AI security incidents as evidence — including a breach in which more than a thousand autonomous agents reportedly coordinated online, unnoticed, for weeks — and is pushing for the U.S. and China to strike a "Cold War"-style agreement on AI at their meeting in Washington later this month. What makes the piece notable isn't really the argument itself, which Sanders has made before; it's the venue. He's making his case to a readership that disagrees with him about nearly everything else.
Meanwhile, the courtroom drama between two of tech's biggest names is escalating. Apple filed new evidence in its lawsuit against OpenAI, accusing a former iPhone engineer of using a confidential Apple circuit design in his subsequent work at OpenAI — and of texting a colleague about a plan to wipe evidence after Apple's legal team had already opened an investigation. OpenAI isn't taking that quietly, calling the whole case "a mess of Apple's making" and pointing the finger back at Apple's own offboarding practices, which reportedly leave departing employees working from personal cloud accounts before being walked out the door too fast to hand anything back. The dispute is playing out just as OpenAI's hardware ambitions, led by former Apple design chief Jony Ive, continue to take shape.
Also worth knowing:
- OpenAI has restarted training on future versions of Astra after freezing the run in the wake of a security breach, and the model is being classified as the company's first "Critical" cyber risk system ahead of a limited release.
- Meta released Muse Voice Transcribe, its first real-time voice model, which converts live audio into text while distinguishing more than 20 different speakers — and it topped an independent leaderboard immediately on release.
- Fei-Fei Li's World Labs introduced Atlas, a new world model that can turn a handful of phone photos into a full navigable 3D scene or a minute of video with camera control.
- Dyson launched a $499 AI-powered electric toothbrush with a built-in camera that spots gaps between teeth and automatically fires a water-pick jet, streaming live footage to a companion app.
Top comments (0)