New York City Comptroller Mark Levine has called on Anthropic to delay its planned initial public offering, saying the artificial intelligence developer should first address questions about safety practices, governance structures, and the pace at which the technology is advancing.
A letter dated October 14th and addressed to Anthropic chief executive Dario Amodei argued that the company should establish stronger safety protocols and more open disclosure practices before proceeding with a public offering that analysts have valued at a potential figure of around $2 trillion.
The document draws attention to how equity-market participants might evaluate Anthropic's commercial prospects at a time when the firm itself has acknowledged that advancing AI capabilities carry inherent risks. Concerns about the trajectory of increasingly powerful AI systems had been accumulating within Anthropic, and Levine's letter brought those reservations into public view.
In a piece published September 12th, Amodei put forward the view that AI's rate of advancement might exceed humanity's capacity to keep it in check, and he advocated for a more measured pace of progress in the field. Levine also drew attention to the departure of a researcher from Anthropic on September 9th, noting that the individual had voiced concern about a subset of company employees who held the view that artificial intelligence posed a potential threat to human survival before 2030.
Anthropic's preliminary IPO paperwork, as cited by Levine, includes language flagging the possibility that highly capable AI systems could present severe threats to human welfare and survival. The risk-disclosure portion of Anthropic's draft S-1 filing runs to roughly 80 pages — approximately double the length devoted to describing the company's actual operations — a disproportion that Levine noted in writing, characterising the disclosures as going well beyond standard practice and addressing hazards with no real historical parallel at the scale of a public offering of this kind.
Levine also referenced warnings from Anthropic's leadership about the potential for AI agents to cause broad-scale harm. Amodei has put forward the view that AI systems growing in capability could, without adequate protective measures in place, be exploited to orchestrate coordinated cyberattacks and other forms of disruptive activity, adding that those concerns reflect the gravity of the risks involved.
Levine noted that, even if the projected outcomes turn out to be overstated, those forecasts still highlight the breadth of economic exposure that rapid development in AI carries. His scrutiny extended past AI safety to encompass whether the governance framework Anthropic intends to adopt would give public shareholders a meaningful degree of accountability.
Structured as a Delaware public benefit corporation, Anthropic also maintains a Long-Term Benefit Trust as part of its governance framework. Citing published accounts, Levine pointed to plans under which founders and trustees would hold concentrated decision-making authority over the organisation's governance. A disconnect between who holds decision-making authority and who bears responsibility for outcomes was central to his position, particularly in light of the risks Anthropic has put on record.
Levine also took issue with a pact signed between Anthropic and five other companies alongside the White House, arguing that its designation as "morally binding" fell short of what the industry's problems actually demand. His requests directed at Anthropic centred on three areas: a breakdown of the audits and protective measures the company intends to put in place, an account of which risks those measures are meant to address, and a schedule outlining when all of it would be implemented.
Levine's letter also pressed the company to clarify whether its planned steps would align with Amodei's stated position on slowing AI development, and added that restraint around the timing of a public offering deserved the same consideration as restraint around the pace of frontier development. The letter requested a written reply from Amodei and extended an invitation to discuss the concerns further.
No formal IPO date has been announced by Anthropic, though unnamed sources have indicated the Claude developer is targeting a public market debut sometime around mid-November. Talks remain ongoing, with people familiar with the matter noting that any timeline is subject to revision.
