- In its amended S-1/A filing Anthropic says internal estimates suggest future models may soon meet its threshold for dramatically accelerated progress attributable to AI.
- No current model has crossed it. The company is considering policy updates to keep humans in control of building the next models.
In short: Anthropic, which is preparing to go public, filed an amended S-1/A filing with the US SEC on 2 Nov 2026. Coverage focused on the financials, but the risk section holds a more important sentence: the company may not notice in time that AI has started to sharply speed up its own development.
What Anthropic actually wrote
- It admits that in the past it did not identify a significant acceleration in its pace of development until after it had happened, because much of its evidence comes from lagging indicators.
- Internal estimates for future models, not current ones, suggest it may soon meet the definition of “dramatically accelerated progress attributable to AI” in its RSP.
- No current model has been designated as exceeding that threshold, and the company is still assessing the underlying data.
- Anthropic writes that it does not expect to be able to safely let Claude fully autonomously design and build successive frontier models without meaningful human supervision. It is considering RSP updates to keep humans in control.
- It calls its advantage “closing the loop”: each Claude version helps build the next, through experiments, training data, evaluations, safety audits and writing code.
Numbers from the same document
- Revenue for the third quarter of 2026: $17.0 billion.
- According to the company, revenue in that quarter grew 1,127% year over year, while compute and infrastructure expenses grew 382%.
How to read this
A prospectus is both a mandatory risk disclosure and a pitch to investors, so a warning about fast progress also signals strength. The idea behind it is recursive self-improvement: AI helping to build better AI. Anthropic does not claim this is already happening at full scale, only that it may soon cross its own warning threshold. Its own wording is cautious, and it promises further reports under the policy. I write about frontier models in the Frontier models hub.