Anthropic's IPO filing says it has met its AI-speeds-up-AI line, on expectation
A day after Claude Mythos 5.2, Anthropic told investors the line is met, but based on future models, not a measured 2×.
On 10 November 2026, the day after releasing Claude Mythos 5.2 (also sold as Claude Fable 5.2), Anthropic amended its stock-market filing (an S-1). It now says it has determined that it has met the “AI research and development capability threshold” in its Responsible Scaling Policy, its published safety rules. The threshold applies where Anthropic observes or expects AI to roughly double the pace of its research.
Zvi Mowshowitz, who quotes the filing, reports that the determination rests “on what we expect from future models” and on the company’s overall rate of progress, not on what it has observed from current models. Anthropic’s August report had said the line was not crossed. METR’s audit of 5 November said it was plausibly crossed by July.
Reviewing Anthropic’s report on the new model, Zvi also cites a point estimate of 1.88× for the speed-up (95% CI 1.62–2.28×), with 32 of 100 refits at 2× or more. We could not confirm which model that figure covers. In its roadshow video, co-founder Tom Brown said Mythos got Claude running on AMD chips in three days with almost no human help, work that took dozens of engineers months before.
What it means
The company with the best view of its own research has now written, in a document investors rely on, that AI is speeding up AI research at about the level its safety policy treats as an early-warning line. It is a statement about expected progress, so it does not show a measured 2× today or AI doing the research alone.
Caveats
- The filing’s wording here comes from Zvi’s quotation. The SEC site refused an automated request, so the filing text was not read directly.
- Anthropic grades itself, in a document written for investors. The 1.88× is as Zvi reports it and was not checked against Anthropic’s system card.
- The estimate sits below 2× and its range spans it. Zvi argues the line is met now and speculates Anthropic held back the released model’s research ability; no source here confirms that.
- The AMD claim comes from a sales video and is not independently checked.