Monday's amended S-1 warned Anthropic might meet its RSI threshold.

Monday's amended S-1 warned Anthropic might meet its RSI threshold. METR's 11-week review puts a number on the acceleration: AECI benchmark progress since October 2025 ran at an estimated 1.9x the earlier rate, with a 1.5-2.8x confidence interval.

METR says faster training-compute scaling did not cause the acceleration. Its best guess is that AI use in R&D accounts for a majority of the speedup, perhaps 80%. That's an uncertain estimate, and the evidence so far points mainly to AI assisting human researchers, though researchers described early signs of autonomous contributions.

I keep coming back to the evidence behind that conclusion. Anthropic redacted a majority of what METR found most informative. METR says the report does not achieve "transparency about evidence," leaving the public to rely heavily on its judgment.

I take METR's finding seriously. But when much of the most informative evidence is redacted, how much of the acceleration can anyone outside Anthropic actually verify?

Sources:
METR: https://metr.org/blog/2026-11-05-anthropic-ai-rd-acceleration-assessment/
METR Substack: https://metr.substack.com/p/2026-11-05-anthropic-ai-rd-acceleration-assessment
Geoffrey Chen: https://geoffreychen.com/ai-technology-observation-2026-11-06-07/

Not financial advice. For informational and educational purposes only.