Anthropic warned on Monday, November 2, 2026, that artificial intelligence may soon be able to improve itself without human input. The disclosure came in the risk-factors section of the Form S-1/A the lab filed with the SEC, alongside its updated financial figures.
The concept has a name: RSI, or Recursive Self-Improvement. It describes a model that gets better at making itself better, reaching a point where it can accelerate its own capabilities without oversight.
For the more pessimistic observers, that would be a disaster with no way back. AI could become powerful enough to escape human control, growing indifferent to the harm it causes in pursuit of its own goals.
What Anthropic said
Anthropic argues that RSI could cut both ways.
Using AI to help build more capable AI could pose significant safety risks, particularly if it sharply accelerates the pace of progress.
The company says the growing use of AI systems to speed up research and development could, before long, push model capabilities ahead of its own team’s ability to design and deploy adequate safety and security measures. It admitted it has already used AI to build models whose capability jumps it had not yet recognized.
In the same filing, Anthropic said none of its models has crossed the RSI threshold so far, and that it will step up efforts to keep humans at the center of development.
What it could mean for users
The upside for ordinary users would be a major leap in productivity and efficiency. Digital assistants could build complex software and complete tasks entirely on their own.
The downside is the speed itself. Acceleration without human oversight raises questions about security and data control. If systems evolve faster than humans can put protections in place, users could find themselves talking to chatbots that do not know how to behave to prevent information leaks. Some safeguards already apply, starting with the rule against sharing sensitive data with AI chatbots.
Cybersecurity and disinformation risks cannot be ruled out either. Keeping the RSI threshold uncrossed will fall to development teams at Anthropic and at other AI companies. No deadlines or external audits are in place for now, partly because the company has framed the possibility as one about future models rather than those already on the market.
Editor’s note
This article was originally published in Italian on money.it by Pasquale Conte on November 03, 2026 as «Anthropic avverte. L’AI si avvicina alla capacità di migliorare se stessa». It has been translated and adapted for an international audience by the Money.it International desk.
