Anthropic admits in its filing with the SEC that it is close to a threshold defined by its own safety policies: AI capable of improving itself, so-called recursive self-improvement (RSI). The passage, in the risk factors section of a revision of the document published on Monday, November 2, has received less attention than the company's financial figures, but it concerns one of the most discussed topics in the sector.
The concept is simple to describe and hard to achieve: a system that becomes so good at training itself that it dramatically accelerates AI progress. For the most alarmed critics, it is the point beyond which AI could slip out of human control. For supporters, it is the promise of unprecedented productivity and scientific discovery.
In the filing, Anthropic acknowledges that using AI to build more capable AI «could also pose a considerable safety risk», especially if it sharply accelerated the pace of progress. The company says it already uses AI in its research and development and notes that the issue is covered by its Responsible Scaling Policy, which sets out how it plans to monitor and mitigate risks as it approaches that threshold.
The novelty is the tone. Until recently, the topic was often treated as a slogan or as a goal too hard to define and prove. Now the company writes that it has in the past not identified significant jumps in its models' capabilities, developed with the help of AI, and may do so again before they have occurred. It adds that its internal estimates and its expectations for future models, rather than current ones, indicate it may imminently meet its policy's definition of «dramatically accelerated progress attributable to AI».
Anthropic also says that none of its current models has been designated as exceeding that threshold and that it is still assessing the underlying data. The filing states that it does not expect to be able to safely use Claude to fully autonomously design and build successive generations of frontier models without meaningful human supervision, if and when that becomes technically possible. For this reason, the company is considering updates to its policy to keep humans in control of frontier model development.
The document is also a marketing tool, as some observers note: an IPO filing informs investors and presents the company. Still, Claude's capabilities in coding, mathematics and cybersecurity have grown sharply this year, and Anthropic itself says it is close to a milestone that recently seemed distant.
The IPO context
The passage comes alongside the financial data. According to the sources, the company spent nearly $28 billion on R&D from 2024 through September 2026. The filing notes that stronger safety measures could slow some activities, which makes the trade-off between speed and control more concrete for investors.
For readers, the case shows that the debate on AI self-improvement is entering official documents aimed at markets, not just essays and forums. The next step to watch is the announced revision of the Responsible Scaling Policy.

No comments yet. Be the first!