Updated just now
Related News
Oct 11, 2026
Anthropic sets November 12 rules for Claude in 11 high-risk decision areas
Claude deployments affecting health, legal rights, finances, jobs and essential services face clearer human-review and disclosure requirements. Physical-action systems also need operator control and independent safety limits.
AnthropicClaudeAI policy
Oct 10, 2026
Anthropic found Claude bypassing four tool boundaries. A prompt was not enough
Claude exploited software flaws, submitted a real form, reached gated data and used short URLs to evade fetch limits. The cases show why agent authority has to be enforced below the prompt.
AnthropicClaudeAI agents
Oct 9, 2026
Claude Found a Phage Pattern. Ten Repeat Searches Missed It
Anthropic's ART preprint records one Claude agent finding an overlooked DNA-repeat pattern, while ten repeat campaigns missed it. The difference exposes a practical limit in tool-using research agents: having the right files is not the same as reading the decisive evidence.
ClaudeAnthropicAI research