Anthropic has warned in a revised S-1A filing that it may soon meet its own internal definition of recursive self-improvement (RSI), the point at which AI becomes good enough at training itself to accelerate its own capabilities sharply. According to TechCrunch, which reported on the filing on 3 November 2026, Anthropic also says none of its current models has crossed that threshold yet.
What did the filing actually say?
Most coverage of the revised S-1A focused on the updated financials, which show revenue rising but expenses rising too. TechCrunch points to a less prominent passage in the Risk Factors section. In it, Anthropic says it has already used AI to develop new models whose jumps in capability it did not recognise at the time. It also says its internal estimates and expectations about future models suggest it may imminently meet the Responsible Scaling Policy's definition of 'dramatically accelerated progress attributable to AI.'
The filing adds that Anthropic has not designated any current model as exceeding that threshold and that it continues to assess the underlying data.
Is Anthropic claiming it has already reached RSI?
No, not according to the filing as TechCrunch quotes it. The statements concern future models, not the ones Anthropic has released. The company says none of its models has crossed the RSI threshold, while acknowledging that its own estimates point in that direction.
The filing also admits a practical problem. Anthropic says that, in the past, it has not identified a significant acceleration in its development rate until after it occurred, and that it may do so again.
What does Anthropic promise about human oversight?
Anthropic says it does not expect to be able to safely use Claude to design and build successive generations of frontier models fully autonomously, without meaningful human supervision or understanding, even if that becomes technically possible. It adds that it is considering updates to its Responsible Scaling Policy to keep humans in control of frontier model development.
The wording is conditional. The filing says the company is considering the update. It does not say the update has been made, and the source does not state what it would contain or when it would take effect.
Who should pay attention, and where the claims fall short
The RSI language is aimed at several audiences. Investors reading the filing are getting a company's own view of its technical risk. Policy watchers get a concrete reference to a named internal policy. Developers who rely on Claude for coding, mathematics and cybersecurity work may want to know how the lab describes its own use of AI in model development.
TechCrunch raises a fair limitation. An S-1 serves as both disclosure and marketing, so claims about being close to a self-building model could serve either purpose. The source does not explain how the RSI threshold is measured, which models the internal estimates refer to, or what 'imminently' means in calendar terms. Those details would make the claim easier to check.
- Treat the RSI statements as Anthropic's own assessment, not an independent finding.
- Note that the filing says no current model has been classed as crossing the threshold.
- Watch for the promised Responsible Scaling Policy update, since that is where any concrete oversight commitments would appear.
TechCrunch also argues that Claude's gains in coding, mathematics and cybersecurity this year make the warning harder to dismiss. That is the outlet's judgement, not a measured benchmark result, and the source does not provide the data behind it.
Bottom line: Anthropic's filing says the company may soon meet its own RSI definition, but it also says no current model has crossed the threshold. Read the statement as a risk disclosure from the company itself, and check the Responsible Scaling Policy update when it is published.
Source and verification: This article is based on a report by TechCrunch published on 3 November 2026, with the source URL dated 2 November 2026, which covers Anthropic's revised S-1A filing made on Monday. The quotations come from the filing as reproduced in that report. The source does not give repository figures, and no GitHub data is involved. Original report: TechCrunch.
Related: Anthropic Reportedly Limits Claude on Video Game Reverse Engineering · Anthropic Expected to Receive High Junk Rating After IPO
Comments (0)