Anthropic Told the SEC Its AI Could Pose an 'Existential Risk.' Within 48 Hours, a Poll and a Federal Probe Said the Public Agrees.
Anthropic's IPO prospectus, detailed by TechCrunch and Fortune on September 28-29, disclosed a 2025 operating loss of roughly $8 billion against $4.6 billion in revenue, $518 billion in future cloud and compute commitments, and a target valuation north of $2 trillion. Buried in the risk-factors section was something no major AI company had put in an SEC filing before: an explicit warning that its own products could pose "existential risks to humanity." TechCrunch called it a first, confirmed by a scan of the SEC's database.
The distinction matters legally, not just rhetorically. CEO Dario Amodei has spent years arguing publicly that frontier AI carries catastrophic risk while continuing to "pace the frontier" of development, a stance Sam Altman and Elon Musk have each echoed in broad terms. But a risk factor in a prospectus bound for a $2 trillion-plus public offering is a different kind of statement: it is a disclosure underwriters vetted and investors are now pricing, not a conference remark. Anthropic becoming the first lab to make that admission in writing, in the exact filing meant to attract public capital, turns a safety-community talking point into a document with legal weight.
The public got there first
A Quinnipiac University poll of 1,202 U.S. adults, fielded September 24-27 and released this week, found 73% of Americans are concerned about an existential threat from future AI systems, 81% say safety should be prioritized over the pace of innovation, and 74% say they distrust the leaders of AI companies to manage the technology responsibly. Ninety-one percent called AI guardrails very or somewhat important, and roughly 80% favored slowing or pausing development for safety assessments.
"More people expect AI to do more harm than good in their day-to-day lives." — Brian O'Neill, Associate Dean, Quinnipiac School of Computing
That the Quinnipiac numbers surfaced within days of Anthropic's own filing is less coincidence than convergence: public skepticism of AI labs' self-governance has been building since similar polling in late September found three in four Americans saying AI firms aren't doing enough to prevent disaster, according to reporting from the Detroit News.
Regulators move in
The third leg landed September 30, when Axios and CNBC reported that FTC Chair Andrew Ferguson is preparing civil investigative demands compelling OpenAI, Anthropic, and other AI companies to turn over documents and testimony on model safety. The probe predates the week's other news, but it crystallizes the same question the prospectus and the poll are both circling: whether the labs building the most capable systems, and the only parties with full visibility into them, can be trusted to self-report the risk honestly. Neither OpenAI nor Anthropic responded to Axios's request for comment, and Ferguson has noted some executives may favor regulation that happens to burden smaller rivals more than themselves — a reminder that converging alarm doesn't guarantee converging incentives.
Three independent signals in under a week — a legal disclosure, a national poll, and a federal inquiry — is an unusual alignment for a debate normally confined to open letters and forecasting tournaments. What it does not yet produce is a binding rule: the FTC's demands are investigative, not enforcement action, and Anthropic's own risk factor is, by SEC design, a liability shield as much as a confession.