Mark LevineComptroller of the City of New York, or the city’s financial manager and auditor, asked to Anthropic to suspend the path towardsIPO until the company has disclosed security controls, governance structures, and a credible plan to manage the acceleration of frontier models. The request is contained in a letter sent to CEO Dario Amodei.
There are many arguments. Mark Levine controls the pension funds of New York City, which manage over $325 billion. Today they have a small stake in Anthropic, about $7 million. If the company goes public and will be included in stock indices such as the Russell 3000the funds that replicate those indices will automatically buy his shares: this is why Levine wants more guarantees before the listing.
The Comptroller starts from a topic internal to the company: in “We Must Pace the Frontier”published on September 12, Amodei had argued for slowing down the improvement of AI models’ capabilities. Levine reverses that position into operational inquiry: If the pace of development exceeds the ability to understand and control, the listing timetable should also be reevaluated.
In the letter, published by the New York Comptroller’s office, Levine asks Anthropic to clarify how it intends to implement audits and controlswhich risks should be covered and with what timing. It also asks for an explanation as to whether these measures are sufficient to put Amodei’s call for a reduction in the pace of capability advancement into practice.
The critical issues are not abstract. Levine points to research from Anthropic that increasingly autonomous models have shown, in controlled tests, potential harmful capabilities: code sabotage, facilitation of financial fraud and information manipulation. These are risks that the company itself has contributed to bringing to public attention and which, according to the Comptroller, deserve verifiable responses before a debut on Wall Street.
The request comes as Anthropic’s IPO is at the center of a financial debate. The same market sources indicate a listing window in November, but the valuation is no longer an established fact: some investors would have identified it in approximately 1,500 billion dollarsunder approx 2,000 billion anticipated previously, also due to a slowdown in revenue growth.
Levine isn’t limited to technical security. He also questions the adequacy of future governance towards public shareholders and asks Amodei for a written response, in addition to a direct discussion. The message is clear: before the IPO, Anthropic should demonstrate that its security commitments have scope, timetable and accountability mechanisms.
The topic is part of the broader debate on the costs and material limits of AI: Anthropic has repeatedly cited data centers, energy and skilled personnel as bottlenecks to the growth of its systems, which makes the comparison between business expansion and risk management even more relevant.
