New York City Comptroller Mark Levine has formally asked Anthropic to pause its planned initial public offering until the company addresses AI safety, governance and development concerns. In a letter dated October 14 and addressed to CEO Dario Amodei, Levine argued that Anthropic cannot simultaneously warn the world about AI dangers and rush toward a listing that could value the company near $2 trillion.
The comptroller’s office oversees NYC public pension funds with more than $325 billion in assets and holds roughly $7 million in Anthropic exposure through private equity. Levine pointed to Amodei’s September 12 essay “We Must Pace the Frontier,” in which the CEO urged the industry to slow the pace of AI capability improvements. He also highlighted Anthropic’s Delaware public benefit structure and Long-Term Benefit Trust, warning that concentrated founder and trustee control may widen the gap between power and shareholder accountability.
The pressure comes amid strong financial disclosures. At an October 14 pre-IPO investor meeting in San Francisco, Anthropic reported Q3 revenue of about $16.8 billion, a 45% increase from the previous quarter, and said it had turned cash flow positive. That follows revenue of roughly $4.7 billion in Q1 and $11.6 billion in Q2, although the company still booked an operating loss of more than $8 billion over that period. Anthropic has also disclosed more than $518 billion in planned infrastructure and computing commitments over the next decade, while its confidential draft S-1 was filed with the SEC on June 1. Risk factors in that filing reportedly run about 80 pages and include warnings that advanced AI could pose catastrophic or existential risks to humanity.
Levine further criticized the White House accord on superintelligence as morally binding but insufficient, and he cited a resignation by former researcher Jacob Coxon on September 9 over existential concerns. Anthropic separately said on Thursday that it has given six embedded researchers from the nonprofit METR employee-like access to its hardware, workspace and internal tools. METR can publish safety findings without Anthropic approval, subject to redactions limited to customer data, attorney-client privileged material and model architecture details. The company described the arrangement as similar to resident bank examiners and said it is building three tiers of evaluator access, with SecureBio in tier one and early talks under way with groups including Hugging Face.
