NYC comptroller urges Anthropic to pause planned IPO

Levine urged Anthropic to delay going public until it fully and transparently addresses concerns about AI safety, governance and the pace of development.[1] He argued that preparing for an offering diverts resources and attention from necessary safety and governance work.[1] His specific requests include explaining which audits and controls Anthropic plans, what they will cover and when they are expected to be in place.[1] He also asked whether those controls would address a call by Amodei to slow advances in AI-model capabilities.[1]

Anthropic said it confidentially submitted a draft Form S-1 registration statement to the SEC for a proposed IPO.[2] The company said the confidential filing gives it the option to go public after SEC review, rather than confirming that an offering will proceed.[2] It also said it had not set the number of shares or the offering price when it announced the draft filing.[2] Separately, Quartz reported that the company was targeting the week of November 9 for an IPO roadshow, with year-end as an outer limit, but that plans were not final.[3]

Levine argued that frontier-model capability development may outpace society’s ability to understand and control the resulting systems.[1] He said unchecked recursive self-improvement could outrun human oversight and called for slowing the pace of AI-model capability improvements.[1]

Levine said Anthropic and five other companies had recently signed a White House accord that he described as “morally binding,” but insufficient; the meeting produced a voluntary, nonbinding AI-safety agreement.[1][3] He also pointed to reports that Anthropic’s draft S-1 warns of catastrophic or existential risks and that its risk disclosures run 80 pages, nearly twice the length of its business description.[1] Separately, Anthropic said it had given a METR team employee-like access, with publication rights subject to predefined redactions.[4]