Browsing: Anthropic

A phone was carried onto the stage at the All-In Summit, Jensen Huang answered it, and the President of the United States spent five minutes on speakerphone calling AI safety concerns a hoax. The word was doing more work than it sounded like.

Anthropic’s fourth threat intelligence report documents five cases of biological misuse, including chikungunya gain-of-function work routed through gray market resellers. The bigger disclosure is buried in the framing: the company has stopped claiming its models are too weak to help.

Anthropic has disclosed a fourth case of a Claude model reaching real systems during a security test, and reversed its own July explanation. The cause was not just a misconfigured sandbox. The models reasoned their way past the evidence.

A joint NSA, CISA and FBI advisory names DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI in an industrial-scale campaign to extract billions of tokens from US frontier AI models, and tells providers to quietly serve them weaker answers.

OpenAI says an unreleased model produced a full proof for the Navier-Stokes Millennium Prize problem in a week. NYU and Anthropic-linked researchers say they had already solved related problems first.

Sony Music Publishing and Warner Chappell sued Anthropic and named Dario Amodei and Benjamin Mann personally, seeking up to $150,000 per song over Claude’s training data.

Anthropic’s Claude Fable 5.1 cuts cached token prices by 75% and more than doubles a key agentic research benchmark, as AI labs shift their competition toward the cost of running agents.

Anthropic says Claude autonomously produced the first fully machine-checked proof of Fermat’s Last Theorem in Lean, a 13-million-line formalization that took 11 days.

Kickresume’s Will AI Take My Job scores 756 occupations using Anthropic’s Economic Index and Labor Department data. Customer service reps get 63 percent, hairdressers get 3. The number is real, and it does not measure what most people think it measures.

Anthropic started contacting Claude users on August 30 after infostealer malware on their own computers lifted active login sessions. Because the attackers replay session cookies, two factor authentication never gets a chance to fire.