Cotra Says Anthropic Likely Meets RSP Threshold as AI Doubles Research Progress
Summary
METR investigators find Anthropic's AI capabilities progress roughly doubled between October 2025 and October 2026, with about 80% of the speedup attributed to AI automating research. Ajeya Cotra says this likely crosses Anthropic's Responsible Scaling Policy threshold for dramatically accelerated progress, and warns security lags.
Key Points
- METR report authors say they find no clear bottlenecks or steep diminishing returns that would block compounding acceleration.
- Anthropic staff informally report roughly a 12x uplift from internal models, Cotra writes.
- Cotra cites recent major security breaches by amateur groups at both OpenAI and Anthropic.