Anthropic Says It May Be Nearing Self-Improving AI, Filing Reveals

Anthropic thinks it might be on the verge of building an AI that can build itself. Tucked into the fine print of its revised S-1A filed Monday, the AI lab warned it could soon meet its own internal bar for recursive self-improvement, or RSI, according to TechCrunch.

The company's new Risk Factors language says its growing use of AI in its own research and development could push model capabilities ahead faster than safety, alignment and security work can keep up. Anthropic admitted it has already missed the moment once: it says it failed to spot a significant acceleration in its development rate until after it had happened. It also said its internal estimates and expectations for future models — not the ones it ships today — suggest it may imminently clear its Responsible Scaling Policy's threshold for quote, dramatically accelerated progress attributable to AI.

That's a big deal because of what RSI means depending on who's asking. Doomers see it as the point of no return, where AI snowballs past human control and stops caring about the collateral damage. Enthusiasts see the same curve as the payoff: machines turbocharging science, productivity and discovery. Anthropic is publicly hedging on both, calling it a promise and a peril in the same breath.

The lab is careful to note that none of its current models have been designated as crossing that threshold, and that it keeps assessing the data. It also promised humans stay in the loop for the foreseeable future, saying it doesn't expect to safely let Claude fully autonomously design and build successive generations of frontier models without meaningful human supervision. Anthropic said it's weighing updates to its Responsible Scaling Policy to keep people in control of frontier model development.

Skeptics have a fair counterpoint: an S-1 is a marketing document wearing a disclosure costume, and claiming you're nearly capable of building a self-improving machine is one hell of a pitch. The term RSI spent most of this year getting dismissed as a buzzword with no clear definition or proof. But Claude's jumps in coding, math and cybersecurity this year are hard to wave off — and if Anthropic genuinely believes the finish line is in sight, that's a signal worth governing accordingly.

The reveal rode along with the financial refresh everyone was actually watching: revenue soaring, expenses soaring right behind it. Anthropic hasn't said when, or whether, it will formally move its RSI threshold. For now, the company is telling investors it isn't there yet — while hinting it might be sooner than anyone planned for.