Anthropic warns it is very close to recursive AI self-improvement, according to its S-1/A

TechCrunch reports that Anthropic, in the revised version of its S-1/A filing submitted on Monday, November 2, 2026, included a striking warning in the fine print: the company may be close to creating a model capable of recursive self-improvement (RSI). While attention has focused on the updated financial figures (revenue is growing strongly, but so are expenses), this passage in the risk factors section has drawn less notice.

According to the article, RSI is the point at which an AI becomes so good at training itself that it triggers a dramatic acceleration of its capabilities. For the most pessimistic, it represents an irreversible cataclysm in which AI builds itself until it escapes human control. For enthusiasts, it is the promise of an unprecedented tool for productivity and scientific discovery. Anthropic appears to see it as both promise and peril.

The filing notes that the company already uses AI in its R&D, but warns that using it to build more capable models “could pose a considerable safety risk, particularly if it sharply accelerates the pace of progress.” It adds that this is covered by its Responsible Scaling Policy, which sets out plans to monitor and mitigate risks as what has been termed “recursive self-improvement” approaches.

There are two new additions to the risk factors included on Monday. First, that Anthropic has already used AI to develop new models whose capability jumps it did not recognize at the time. Second, that it could soon meet its own definition of RSI. The text states that the growing use of AI systems to accelerate its research may cause capabilities to advance faster than its ability to develop, validate and deploy safety, alignment and security measures. As a result, in the past it did not identify a significant acceleration until after it had occurred, and that could happen again.

According to the quoted passage, internal estimates and expectations about future models, not current ones, indicate that the company could “imminently” meet its policy's definition of “dramatically accelerated progress attributable to AI.” Anthropic clarifies that it has not designated any of its current models as above that threshold and that it continues to evaluate the underlying data.

The company also pledges to keep humans in the R&D process for the foreseeable future. It states that 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 oversight or understanding, should that become technically possible. It is therefore considering updating its Responsible Scaling Policy to keep people in charge of frontier model development.

The article also includes a skeptical caveat: an S-1 is both a disclosure document and a marketing document, and claiming to be close to a model capable of building itself could fall into the latter category. Even so, the author, Julie Bort, argues that Claude's capabilities, especially in coding, math and cybersecurity, have grown enormously this year. Her conclusion is that if Anthropic believes RSI is near, it may be wise to take it at its word and act —and even regulate— accordingly.

Note: this summary is based solely on the text of the article received, which appears to be complete; no external data has been added.