METR finds faster AI research at Anthropic, with important evidence still private
METR estimates that growth in Anthropic’s internal capability index accelerated 1.9 times, but the cause and the most informative supporting evidence remain partly uncertain or private.
METR’s November 5 assessment puts a measured range around one part of the AI research feedback loop.
Growth in Anthropic’s internal index of well-specified capabilities accelerated 1.9 times relative to its earlier trend, with a 95% confidence interval of 1.5–2.8. This is a comparison of capability-index growth rates, not a finding that individual researchers became 1.9 times more productive. [1]
The evaluator judges that AI assistance probably explains most of the current acceleration, while acknowledging other organizational changes and uncertainty about causation. Anthropic provided extensive internal access between August 12 and October 29, but redacted much of the evidence METR considered most informative. The public report therefore supports a serious acceleration signal without allowing outsiders to reconstruct its strongest arguments. [1]
Sources & details 2 sources
Discuss in your voice app
Take the reporting, visual explanations and original sources into a conversation. Copy the brief, paste it into your preferred assistant, then start its voice mode.
Paste the brief into your voice app to discuss this reporting.Reporting sources
- METR’s acceleration assessment
November 5 report; internal-index estimate, causal judgment and redaction limits. Public evidence does not independently reproduce the private measurements.
- METR’s review of Anthropic’s August risk report
Separate November 5 assessment of the early-warning threshold. Same evaluation organization, not independent replication.
Images & figures
METR estimates 1.9 times the earlier rate of growth in Anthropic’s internal capability index; the 95% confidence interval is 1.5–2.8.
Euangelion · Source · Usage terms · Download figure