Anthropic has told prospective investors that it may soon reach a point where its AI models dramatically speed up the company's own research. In a revised IPO filing submitted on November 2, 2026, the Claude developer added risk factor language saying its internal estimates indicate it "may imminently meet" the threshold its Responsible Scaling Policy sets for accelerated progress attributable to AI, according to TechCrunch. (TechCrunch (opens in a new tab))
The idea behind the warning is known as recursive self-improvement, or RSI: the point at which AI becomes capable enough at building AI that progress sharply accelerates. TechCrunch noted that the term gained wide use earlier in 2026, and that people read it in very different ways. Some see it as a point of no return where AI could slip out of human control, while others see a path to large gains in productivity and science. (TechCrunch (opens in a new tab))
What the filing says
The disclosure sits in the Risk Factors section of the amended S-1, according to International Business Times Singapore. Anthropic writes that it uses AI in its own research and development, and that this could cause its models' capabilities to advance faster than its ability to develop matching safety, alignment and security measures. (International Business Times Singapore (opens in a new tab))
The company states plainly that "using AI to help build more capable AI could also pose a considerable safety risk." It also concedes a detection problem. According to TechCrunch, the filing says Anthropic has in the past failed to identify a significant acceleration, and may fail to identify one in the future. IBTimes reported that the company says it did not recognize such a speedup until after it had already happened. (TechCrunch (opens in a new tab); International Business Times Singapore (opens in a new tab))
Anthropic's Responsible Scaling Policy uses a concrete benchmark for this kind of acceleration, IBTimes reported: a doubling of the baseline rate of improvement that can plausibly be credited to AI. Crossing it is meant to trigger additional safeguards. The company says it has not designated any of its current models as exceeding that threshold, but its expectations for upcoming models suggest that could change soon. (TechCrunch (opens in a new tab); International Business Times Singapore (opens in a new tab))
Humans kept in the loop
The filing also draws a line on autonomy. Anthropic says it does not anticipate being able to safely use Claude to fully autonomously design and build its successive frontier models without meaningful human supervision or understanding. It expects to keep people involved in its research for the foreseeable future and says it is considering updates to its Responsible Scaling Policy to keep humans in control of frontier model development, according to both outlets. (TechCrunch (opens in a new tab); International Business Times Singapore (opens in a new tab))
Both reports added a caveat. An S-1 is written for investors, and TechCrunch pointed out that such a document is partly a marketing tool. IBTimes made the same point, noting that a warning about fast progress can also signal to buyers of the stock how capable the company's models have become. (TechCrunch (opens in a new tab); International Business Times Singapore (opens in a new tab))
The disclosure lands in an unusual place. Safety researchers have discussed recursive self-improvement for years, but it is rare for a company to describe the risk in a legal document that carries liability if it is misleading. By putting the language in its prospectus, Anthropic is telling investors two things at once: that its technology may be close to a major leap, and that the leap could create risks the company might not spot in time. Whether the threshold is crossed, and how Anthropic would announce it, are now questions with financial as well as safety weight.



