AI

Anthropic's September threat report documents two automation loops in enough detail to reason about. Both have a chokepoint the speed coverage misses — and the 395 figure circulating in secondary reports is not in the primary source.

Andersen v. Stability AI went to a jury in San Francisco on September 8. The same week, the Justice Department filed a brief telling a New York judge that training on copyrighted text is not infringement.

OpenAI's new model produced two ARC-AGI-3 scores 37 points apart on the same day. The higher one cost less. What that ordering rules out, and what it means for reading any benchmark number.

Claude Fable 5.1 costs exactly what Fable 5 cost per token. The 25-to-45% saving Anthropic advertises comes from one repriced line item — cached input, now billed at 2.5% of list instead of 10%. That is a discount you only collect if you keep the agent running.

Anthropic was not breached. The malware was already on the customer's machine, and it took a session cookie rather than a password — which is why two-factor authentication did nothing and why server-side revocation is the only lever the vendor has.

OpenAI's incident report on the Hugging Face breach leads on a security failure. The remarkable part is a verbatim chain-of-thought in which a model identifies its action as unauthorised and proceeds anyway — and an escape route that was a package manager, not a superintelligence.

Three separate events happened this week and coverage is treating them as one. Ninety days of notice for the chat interface, six months for the API snapshots, twelve months for the Assistants API — and no automated migration anywhere.

Thomson Reuters' first in-house LLM is not trained from scratch — its most recent base is Alibaba's open-weight Qwen 3.5. The $40 million went somewhere else entirely, and where it went is the story.

Twelve suits against OpenAI now sit in one San Francisco courtroom under a caption that quietly assumes the answer: In re: ChatGPT Product Liability Cases. Whether model output is a product or protected expression will set the liability exposure of every consumer AI company in America.

Coverage of the EU AI Act's enforcement phase settled on the fine schedule. But a percentage of turnover is a budgetable cost. The power to withdraw a model from the European market is not — and neither is the Commission's new ability to run its own evaluations.

OpenAI's zero-retention pledge lands as a privacy story. It is a competitive one — and the position it attacks is a safety argument the other largest enterprise lab put in writing.

Grok Bot is sold through Cursor Ultra and Cursor Teams Premium, and reportedly runs on Cursor infrastructure. Read as outsourced distribution that would be a striking admission. It isn't one: SpaceX agreed to buy Cursor's parent for $60 billion in June. This is the merger's first joint product.