METR Audit Says Anthropic Model Acceleration May Have Triggered Warning Threshold
Understand the story
Between August and October 2026, METR conducted an external audit of Anthropic, finding that since October 2025, the improvement speed of its model test scores had been roughly 1.9 times the previous rate: AECI annual growth rose from 15 points to 29 points. METR estimated that about 80% of the speedup came from AI being used in its own research. The audit also concluded that Anthropic may have reached the acceleration warning threshold set by the company itself as early as July 15, 2026. Currently, the only public information is the disclosure of the audit's conclusions; no response from Anthropic has been reported yet.
AI overview of reporting · Updated 2 h前
Reporting timeline
Explore different aspects of the story through its reporting.
- The DecoderMETR finds Anthropic may have crossed its own AI acceleration warning threshold back in July
METR conducted an external audit of Anthropic from August to October 2026, finding that since October 2025 its models' test scores improved about 1.9x faster than before (AECI annual growth rising from 15 to 29 points), estimated that roughly 80% of the speedup came from AI used in its own research, and that the company's acceleration warning threshold may have been reached as early as July 15.
Story heat trend
Not enough continuous observations to show a trend yet.