Anthropic’s S-1/A on RSI: falsify the Risk Factors wording, not “self-building already”

While markets watched soaring revenue and costs in Anthropic’s amended S-1/A, TechCrunch’s Julie Bort (2 Nov 2026) focused on new Risk Factors sentences: internal estimates and expectations for future models indicate the company “may imminently” meet its Responsible Scaling Policy (RSP) definition of “dramatically accelerated progress attributable to AI.”

Falsify the text. If later official RSP updates, safety reports, or regulatory Q&A show a current model formally designated past that threshold, or show the “imminent” claim walked back with no governance follow-through, revise today’s weight on this risk factor. Treating S-1 wording as proof that “self-building AI has landed” goes beyond what the sentences support.

Hard facts (TechCrunch’s account of the S-1/A, 2 Nov 2026; filing submitted to the SEC the same day):

• Risk language: Anthropic says rising use of AI to speed its own R&D may advance model capability faster than matching safety, alignment, and security measures can be developed, validated, and fielded; it has in the past failed to spot a large acceleration until after the fact, and may again.
• Threshold phrasing: internal estimates and expectations about future models, explicitly contrasted with current models, indicate the firm “may imminently” meet the RSP definition of “dramatically accelerated progress attributable to AI.”
• Present-tense caveat: no current model is designated as exceeding that threshold; underlying data remain under assessment.
• Humans in the loop: the company says it does not anticipate being able to safely use Claude to fully autonomously design and build successive frontier generations without meaningful human supervision or understanding, if and when that becomes technically possible; it is therefore considering RSP updates to keep humans in control of frontier model development at Anthropic.
• Frame: the story notes an S-1 is both disclosure and marketing; Claude’s visible gains in coding, math, and cybersecurity that year are background, not proof the threshold was crossed.

Evidence strength: secondary reporting of a primary securities filing, plus a public S-1/A URL. TechCrunch prints long direct quotes; this piece keeps “may imminently meet the RSP definition,” “no current model past the line,” and “considering stronger human control” in one evidence bundle so the headline cannot travel alone.

Steelman: putting nearness to an RSP acceleration threshold into Risk Factors is a compliance signal to investors and a named governance object for the safety community: moving “AI speeds AI” from slogan to something that can be cited. Keeping humans in the loop in the same passage reduces a runaway reading. If management believes the threshold is near, writing it into the prospectus before the fact fits disclosure logic better than a patch afterward.

Strongest steelman: Risk Factors must serve counsel, investors, and the public at once, so the language stays hedged; “may imminently,” “future models,” and “still assessing” already limit certainty.

Steelman risks: the same text can read as trailer-style marketing: using near-RSI narrative to heighten scarcity. The RSP threshold’s operational definition, metrics, and independent audit path are not unpacked in the story. Admitting past failure to spot acceleration on time also undercuts confidence that a crossing moment can be crisply announced. Readers lack a public time series that would let outsiders verify “imminent.”

For investors and analysts, watch whether later RSP text, model cards, or safety reports publish checkable acceleration metrics, and how “human supervision” moves from principle into process. For policy readers, watch whether securities disclosure becomes a harder channel for AI-risk communication than blog posts. For competitors and customers, the six-month check is whether any model is formally labeled past the threshold and whether autonomous R&D boundaries enter external policy, not social-media slogans about “self-building AI.”

Do not read TechCrunch’s “very close to RSI” headline as proof RSI already exists. The falsifiable focus is Anthropic’s Risk Factors sentence cluster: whether future-model expectations trigger the RSP “dramatically accelerated progress attributable to AI” definition, whether current models remain undesignated, and whether human-control language lands in an updated RSP. An S-1 is disclosure plus marketing; governance moves and public metrics are what to check next.

【No.2】