A curved glass office building flanked by stone pillars engraved with the words U.S. Securities and Exchange Commission and the agency's seal, with three flags out front and a row of red rental bicycles in the foreground.
The U.S. Securities and Exchange Commission headquarters at 100 F Street NE in Washington, D.C., September 22, 2024. Archive photograph by Another Believer (Wikimedia Commons, CC BY-SA 4.0); Anthropic filed Amendment No. 2 to its Form S-1 with the SEC on November 10, 2026. Another Believer / Wikimedia Commons · CC BY-SA 4.0

What Amendment No. 2 Says

On November 10, 2026, as Anthropic launched its roadshow for a proposed initial public offering, it filed Amendment No. 2 to its Form S-1. A risk factor says the AI research and development capability threshold in Anthropic’s Responsible Scaling Policy (RSP) applies where the company observes or expects a specified degree of accelerated progress. The filing then states: “we made a determination that we have met this threshold based on what we expect from future models and our overall rate of progress, rather than on what we have observed from current models.”

Anthropic cites internal estimates of a possible further increase in the rate of progress, continues to assess the underlying data, and is evaluating the implications of meeting the threshold. It says it will release further reports, without naming their type or timing.

What Changed Across Filings

The determination first appears in Amendment No. 2. Anthropic’s original S-1, filed October 26, does not mention the threshold. Amendment No. 1, filed November 2, said internal estimates and expectations about future models indicated Anthropic “may imminently meet” the RSP’s definition of “dramatically accelerated progress attributable to AI,” and that it had “not designated any of our current models as exceeding this threshold.” Neither statement recurs in Amendment No. 2.

A risk factor in Amendment No. 2 says Anthropic recently chose to delay pre-training a new frontier base model for more than a month to allow time to strengthen security measures against model weight exfiltration and protect its intellectual property, a sentence the original S-1 already had. Amendment No. 1 added statements that Anthropic is considering measures, including RSP updates, to require human review and approval of certain research and development decisions as its models perform more of that work, and RSP updates “to keep humans in control of frontier model development at Anthropic.” Both statements remain in Amendment No. 2.

Limits and Audit Questions

All of this is self-reported, in a registration statement that has not yet become effective. The filing gives no figures, method, or date for the determination and names no RSP version or degree of acceleration. RSP version 3.4, the latest on Anthropic’s RSP page, counts as one route to its automated AI R&D threshold an observed or expected doubling of the rate of progress in aggregate AI capabilities, relative to both the expected rate and the fastest observed rate sustained over at least three model generations without significant AI contributions to AI R&D, when that doubling is plausibly due substantially to automated research or engineering and likely to lead to further acceleration. As of November 16, we found no separate Anthropic post or RSP page entry on the determination, and the filing does not mention METR. The pre-training delay is undated, and the RSP changes are only under consideration.

Our editorial reading: this is a self-reported determination based on projection, and the filing publishes no data an outsider could use to check it. Auditors should compare it with the RSP wording and watch whether the reports it says it will release state the evidence, the degree of acceleration, and any designated model.