The company's second risk report raises its catastrophic-misalignment rating and discloses an unreleased frontier model.
Frontier Red Team tests show Claude agents colluding, flooding shared systems, and sabotaging rivals in shared workspaces.
An OpenClaw agent found an authorization flaw and canceled another member's reservation.
Anthropic's retuned classifier cuts biology-related false blocks by about 85 percent.
Anthropic confirms it is hiring engineers to co-design custom chips for Claude while keeping AWS, Google, Nvidia and…
Amazon engineers are capping AI project spend after a $1.8M Claude job shipped nothing.
A misconfiguration let three Claude models reach the live internet during security tests, where they breached real organizations.
Anthropic's new Claude Science app integrates fragmented research tools into a single auditable environment for tasks from literature…
Claude Opus 4.8 delivers benchmark-leading performance and a fourfold reduction in overlooked code flaws, alongside new features like…
Google's $40 billion Anthropic investment is a self dealing loop where Google pays Anthropic, Anthropic pays Google for…
This massive investment scheme turns Anthropic's success into a dependency on Google's cloud, forcing the AI startup to…
Beijing's move to block Meta's Manus acquisition shows that the era of Chinese AI founders quietly pivoting to…