Key facts
- Anthropic said none of its current models has been designated as exceeding its recursive self-improvement threshold.
- Anthropic said its internal estimates and expectations for future models point to the threshold being met imminently.
- The filing stated Anthropic has in the past not identified an acceleration in its rate of development until after it occurred.
- Anthropic said it is considering updates to its Responsible Scaling Policy to keep humans in control of frontier model development.
- The report said the revised S-1A showed revenue rising while expenses also rose.
Anthropic said in new language added to the risk factors of its revised S-1A registration statement that its internal estimates and its expectations for future models indicate it may imminently meet its own Responsible Scaling Policy definition of “dramatically accelerated progress attributable to AI,” the company said on Monday, in a report that placed the filing on Monday without a calendar date. The filing also cautioned that its growing use of AI systems to speed its own research and development may let model capabilities advance faster than its ability to develop, validate and implement safety, alignment and security measures.
Anthropic said it has in the past failed to identify a significant acceleration in its rate of development until after it had occurred, and may fail to identify one in the future. It also said it has not designated any current model as exceeding the threshold and that it continues to assess the underlying data.
The company said it does 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. It said it is considering updates to its Responsible Scaling Policy to keep humans in control of frontier model development at Anthropic.
Recursive self-improvement describes the point at which an AI system becomes capable of training itself well enough to accelerate its own capabilities sharply. The report described the term as having been treated by many as a buzzword, or at best a goal too hard to achieve, define or prove, when it gained popularity this year.
The report said Anthropic's revised filing also disclosed updated financials, with revenue rising and expenses rising too. It said the company's models have shown large gains in coding, math and cybersecurity this year, an assessment it attributes to Anthropic's own view of Claude.