Anthropic filing says it may soon meet its own threshold for AI-accelerated progress

Anthropic has added language to its revised S-1A filing warning that it may soon meet an internal threshold for AI speeding up its own development, according to TechCrunch.

The filing was revealed Monday. TechCrunch said most attention went to its updated financials, which show revenue soaring along with expenses. But the outlet pointed to new statements in the Risk Factors section.

In them, the company says its growing use of AI systems in its own research and development may cause its models to advance faster than it can develop, validate and put in place matching safety, alignment and security measures. The filing says the company has "in the past not identified, and may in the future not identify, a significant acceleration in our rate of development until after it has occurred."

It adds that internal estimates and expectations about future models, rather than current ones, indicate the company "may imminently meet" its Responsible Scaling Policy's definition of "dramatically accelerated progress attributable to AI."

TechCrunch described this as a warning that Anthropic is close to recursive self-improvement, or RSI. The outlet defined RSI as the point at which AI becomes so good at training itself that it causes a dramatic acceleration in AI capabilities. Some see that as a path to AI slipping beyond human control, TechCrunch noted, while others see it as a source of huge gains in productivity and scientific discovery.

The filing itself names the risk. It says that "Using AI to help build more capable AI could also pose a considerable safety risk, particularly if it sharply accelerates the pace of progress."

A lot is still unclear. The filing says Anthropic has not designated any of its current models as exceeding the threshold, and that it continues to assess the underlying data. TechCrunch's report does not say which earlier models showed jumps in capability that the company did not recognize at the time.

Anthropic also says it is considering changes to its Responsible Scaling Policy, but the report does not describe what those changes would be. The filing states:

“We do not anticipate being able to safely use Claude to fully autonomously design and build successive generations of frontier models without meaningful human supervision or understanding, if and when that becomes technically possible. Accordingly, we are considering updates to our Responsible Scaling Policy to keep humans in control of frontier model development at Anthropic.”Anthropic, revised S-1A filing

TechCrunch also noted that skeptics might argue an S-1 works as both disclosure and marketing. In that view, saying a self-building model is close could be part of the pitch.

RM Roger Morais Editor Roger Morais is a Portuguese-American journalist and web developer who founded RogerLe.com in 1996. He spent ten years at the Federal Reserve Bank of Dallas, where he was a senior web developer, and now travels Europe full-time by motorhome. He decides what morais.news covers and approves every piece before it is published.