Anthropic redacted parts of the 11-week review, leaving gaps in the public record on frontier AI development.
METR published a report on Thursday after an 11-week review inside Anthropic and found that the company’s AI models are improving roughly twice as quickly as they did a year earlier. Benchmark scores suggest Anthropic made about two years of historical progress in the past year. METR researchers concluded that scaling compute alone does not explain the acceleration and that AI now drives most of it.
Anthropic agreed to the METR review, which covered its AI research and development work. The company could redact portions of the final report under that agreement and used the option broadly. The review is separate from a program that gave safety researchers employee-like access to systems after hacking incidents. Anthropic is also preparing to go public while CEO Dario Amodei calls for slower AI development.
Anthropic employees using the company’s most capable internal model reported about 12 times the productivity of working without AI, according to METR. That self-reported gain matters for operators because internal AI tools can compress research cycles and shift capacity toward oversight, evaluation, and security work. Teams should plan for fast release cadences and treat redacted safety evidence as a governance risk.
The redacted sections leave outside teams with limited evidence for judging Anthropic’s safety claims and the pace of AI-driven research gains. Anthropic’s preparation for a public offering raises the stakes for clear disclosure and independent evaluation. Watch whether Anthropic releases redaction detail, frames self-improvement risk openly, and enables outside evaluation of compute scaling versus AI contribution.
What matters
- METR found Anthropic models improving roughly twice as fast as they did a year earlier.
- Operators should treat AI-driven research gains as a planning variable, not a fixed baseline.
- Watch whether Anthropic discloses the redacted sections as its public offering plans advance.
Why it matters
Watch whether Anthropic discloses the redacted sections as its public offering plans advance.
This GenAI News article was prepared in original wording using reporting and materials published by The Verge. Source reference: https://www.theverge.com/ai-artificial-intelligence/1029126/anthropic-metr-report-accelerating-ai-development.
Drafted by the GenAI News review pipeline.