METR Audit Finds Anthropic AI Models Accelerating Faster With AI Assistance
First merged summary.
METR audit shows Anthropic models improving 1.9 times faster via AI assistance, with warning threshold potentially crossed in July.
METR audited Anthropic from August to October 2026 and found its models progressing 1.9 times faster than a year earlier, with about 80% of the gains from AI assisting researchers rather than added compute. The auditors identified no bottlenecks that would halt further acceleration and assessed low acute catastrophic risk. Anthropic applied extensive redactions to the report and trained three new unreleased models in the past six months.
- Models gained 29 AECI points per year since October 2025
- Internal self-assessments report 12x productivity gains
- Majority of gains from AI in research not compute
- Three new unreleased models trained in past six months
- Low acute catastrophic risk
- Heavy redactions by Anthropic
What each outlet adds
beyond the shared facts- The DecoderAI acceleration warning threshold potentially crossed in July, 29 AECI points per year, 80% gains from AI research
- The Vergemodels improve twice as fast, three new unreleased models trained
Coverage
oldest first · one row per article · UTC- · 5d ago
The DecoderFirst report
METR finds Anthropic may have crossed its own AI acceleration warning threshold back in July
METR's audit of Anthropic reveals models progressing 1.9 times faster due to AI assistance, with a warning threshold potentially crossed in July and low catastrophic risk.
- · 4d ago
The Verge · AI
METR report finds Anthropic AI accelerating own development
Models improve twice as fast via AI assistance with heavy redactions limiting public insight.