Anthropic Says Recursive Self-Improvement May Be Imminent

Anthropic says it may be close to the point its own safety rules treat as the start of recursive self-improvement, the moment AI begins to speed up AI research on its own. The warning appears in the amended S-1 filing the Claude maker sent to the SEC on November 2, ahead of its planned Nasdaq listing.

The key sentence is new. It was not in the first draft of the filing from October 26: “Our internal estimates and our expectations regarding future models, rather than our current models, indicate that we may imminently meet our Responsible Scaling Policy’s definition of ‘dramatically accelerated progress attributable to AI.'”

What Recursive Self-Improvement Means for Anthropic

Anthropic does not claim it has crossed the line. The filing says the company has “not designated any of our current models as exceeding this threshold” and is still assessing the data. The forecast is about models that have not been released yet.

The bar comes from the company’s Responsible Scaling Policy, now at version 3.4. In plain terms, it means AI doubling the rate of progress in overall AI capability, so that one year brings what would normally take two. Elsewhere in the S-1/A, Anthropic links this area of its policy directly to what “has been termed ‘recursive self-improvement.'”

The filing also shows how much Claude already does inside the company. Over 90% of contributions to Anthropic’s codebase in September 2026 were written by Claude, and the company calls the cycle of each model helping build the next one “closing the loop.” Tools like Claude Code are part of that internal workflow.

Anthropic Admits It May Notice Too Late

The most striking part is about detection. Anthropic says much of its evidence on AI-driven acceleration comes from lagging indicators, and that it 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.”

In response, the company says it is considering updates to its scaling policy to require human review and approval of certain research decisions as models take on more of that work. It also says it does not expect to safely let Claude design and build new frontier models “without meaningful human supervision or understanding.” The filing adds that Anthropic recently delayed pre-training a new frontier base model by more than a month to tighten security around model weights.

Why It Matters Before the IPO

An S-1 risk factor is both a legal shield and a pitch to investors, a point Gizmodo raised after TechCrunch first spotted the change. Still, it is rare for a frontier lab to put an “imminent” capability threshold in writing to regulators.

Rivals are racing in the same direction, with models like GPT-6 Astra already editing other models’ weights. For the full listing timeline, see our Anthropic IPO guide.