Anthropic's S-1A says it may imminently meet its own threshold for AI-accelerated progress

Anthropic's revised S-1A, filed Monday, includes new language in its Risk Factors section saying the company may soon meet its own internal definition of accelerated, AI-driven progress. Most attention to the filing has gone to the updated financials, which show soaring revenue alongside rising expenses, but TechCrunch flagged the risk disclosure as a significant addition.

The filing says Anthropic's internal estimates and its expectations for future models, rather than its current ones, indicate it 'may imminently meet' the Responsible Scaling Policy definition of 'dramatically accelerated progress attributable to AI.' It adds that it has not designated any current model as exceeding this threshold and continues to assess the underlying data.

Anthropic also states that its growing use of AI in its own research and development may make model capabilities advance faster than its ability to develop, validate and implement safety, alignment and security measures. As a result, it says, it has in the past not identified a significant acceleration in its rate of development until after it happened, and may fail to do so again.

Recursive self-improvement is viewed very differently depending on who is asked. To AI doomers, TechCrunch writes, it is the point beyond which AI could escape human control. To enthusiasts, some version of AI reliably improving itself represents the technology's promise in productivity and scientific discovery. Anthropic's filing treats it as both a promise and a peril.

The disclosure also gives a concrete, if hedged, marker for a concept that many people dismissed as a buzzword or as too hard to define or prove. The filing says Anthropic does not expect to safely use Claude to fully autonomously design and build successive generations of frontier models without meaningful human supervision, and that it is considering policy updates to keep humans in control.

TechCrunch's own caveat is worth noting: an S-1 serves as both disclosure and marketing, and claims of being close to a model that can build itself could fall into the latter category. The outlet nonetheless points to Claude's gains this year in coding, math and cybersecurity as a reason to take the warning seriously.

The main open item is whether Anthropic follows through on the Responsible Scaling Policy updates it says it is considering. The filing does not give a timeline or detail what the changes would require.

Also unresolved is the assessment itself: Anthropic says it continues to assess the underlying data on whether its models have crossed the threshold, so any later designation of a model as exceeding it would be a notable development.