Key takeaways
- Anthropic's November 2 amended S-1 says its estimates indicate it may imminently meet its RSP definition of dramatically accelerated progress attributable to AI.
- It says no current model has been designated as exceeding that threshold and it is still assessing the data.
- Anthropic says it is considering Responsible Scaling Policy updates to keep humans in control of frontier model development.
Anthropic says in its amended S-1 filing that its own estimates indicate it "may imminently meet" the company's definition of "dramatically accelerated progress attributable to AI". The risk factor, in the filing with the SEC on November 2, is the clearest statement yet from a frontier lab that AI is speeding up its own development. TechCrunch headlined it as a warning that Anthropic is close to recursive self-improvement, or RSI.
RSI means AI systems getting good enough at building AI that progress speeds up sharply. The filing doesn't claim that has happened. It says the risk is real enough to disclose to investors.
What the filing says
Anthropic calls its practice "closing the loop": successive versions of Claude help build the next one. The filing says Claude models are "increasingly becoming critical tools" in designing, training, testing and optimizing the next generation.
The risk factor lists what could go wrong:
- Capabilities could advance faster than safety, alignment and security measures.
- A flaw in one model could carry into the next and be hard to detect.
- Anthropic's evidence about the pace of acceleration comes from "lagging indicators", so it has in the past not noticed a significant acceleration until after it happened.
The threshold it mentions comes from its Responsible Scaling Policy, the company's own framework for when to add safeguards. Anthropic says it has not designated any current model as exceeding the threshold and is still assessing the data. It says it will release further reports under the policy.
What Anthropic says it will do
According to TechCrunch's reading of the filing, Anthropic says it does not expect to safely use Claude to design and build successive frontier models fully on its own, without meaningful human supervision. It is considering updates to its Responsible Scaling Policy to keep humans in control of frontier model development.
How much to read into it
An S-1 is a legal disclosure, written to list every risk an investor might hold against the company later. TechCrunch notes skeptics who say it is also a marketing document, and that claims of nearly self-building AI could be promotional. The filing gives no measurement of how much faster development has become.
Our earlier article covers the financial numbers in the same filing. Watch for the Responsible Scaling Policy reports Anthropic says are coming.