Anthropic’s threat chief told CNBC that Moonshot’s Kimi K3 was illegally trained off Claude — the same day Reuters sources said the lab confidentially filed a Hong Kong IPO.
Anthropic shipped Claude Fable 5.1 and Mythos 5.1 today. Same model, two safeguard paths, 75% cheaper cache reads, and a privacy stack that keeps monitoring data on your cloud.
Sony Music Publishing and Warner Chappell filed a federal copyright suit in Northern California accusing Anthropic and two co-founders of torrenting and scraping musical compositions to train Claude. Anthropic says it disagrees and will defend itself in court.
Anthropic opened a research preview of the Model Hardware Standard — a shared driver so AI agents can run lab and factory gear. It started with HHMI Janelia. Open source comes later, after safety work.
On 27 August OpenAI posted an open letter asking for a surge in cyber defense. TechCrunch, CNBC, and the BBC say more than 100 companies signed it, including OpenAI, Anthropic, Google, and Microsoft. I am treating the letter as the story, not yesterday’s Hugging Face paper.
Thursday night a San Francisco judge said the Pentagon’s supply-chain-risk label on Anthropic was illegal retaliation, not a real sabotage case. That is one court. A second case in D.C. is still open, so the label is not cleanly gone.
Anthropic opened Claude in Chrome to every paid plan today, and Claude can take browser actions without asking every time. Here’s the install, the three permission modes, and why I’d still keep it off my bank tab.