Frontier Risk Report (February to March 2026)

Gives rare third-party evidence on how internal frontier agents are monitored and how capable they are, drawn from non-public lab data, which informs how far to trust autonomous agents.

• article前沿 AI 风险报告（2026 年 2–3 月）metr.org
• articleRed-Teaming Anthropic's Internal Agent Monitoring Systemsmetr.org
• articleReview of the "Risks from automated R&D" section in the Anthropic Risk Report (February 2026)metr.org

Checking sign-in…

Loading comments…