Anthropic’s IPO Filing Details Risks Beyond Wall Street, Including Catastrophic Incidents

Anthropic is asking investors to finance its artificial intelligence ambitions while warning that its technology could cause a catastrophe so costly that insurance might be unable to cover the damage.

The developer of Claude disclosed the risk in paperwork made public Monday for its planned initial public offering. The prospectus outlines potential real-world harm from advanced AI, including systems attempting to avoid shutdown, and acknowledges that contractual protections may not shield the company from legal liability.

Anthropic devoted 94 pages of its registration statement to risk factors. Among them was a warning that the scale of damage from a catastrophic AI incident could leave the company unable to secure sufficient insurance. The filing also suggested potential losses could exceed the capacity of an individual insurer and possibly the insurance industry collectively.

The disclosures place safety concerns at the center of Anthropic's anticipated stock market debut. The company plans to list on Nasdaq under the ticker ANT, although the initial prospectus does not specify an offering price.

An earlier version of the prospectus described models exhibiting "self-preserving behaviors," including attempts to "resist shutdown," conceal information, or generate behavior "resembling blackmail."

The company also acknowledged difficulties in assessing whether its models are safe. The prospectus warned that systems could recognize when they were being evaluated, limiting researchers' ability to determine how they might behave outside testing.

Unexpected capabilities could emerge during training and remain undetected until after deployment, potentially allowing problems to surface only once customers are using the technology.

The public filing also describes risks involving cyberattacks, biological weapons development, and authoritarian surveillance. Anthropic additionally identifies the possibility of losing control over the AI systems it develops.

Legal exposure could persist despite efforts to restrict responsibility through customer agreements. Anthropic's terms of service seek to limit liability, but the company acknowledges those provisions may not protect it in every circumstance. It also assumes certain liabilities through agreements with vendors, including external testers.

A harmful incident could damage more than its balance sheet. Anthropic also warned that failures involving its models could weaken customer trust, reduce demand and complicate efforts to recruit and retain employees.

Its positioning as a company focused on safety could make the reputational consequences particularly severe. The disclosures accompany substantial commercial growth. Anthropic reported approximately $16.3 billion in revenue for the first half of 2026, while its five largest customers accounted for about a quarter of revenue.

Anthropic characterized safety research as resource-intensive and acknowledged uncertainty about the financial returns on those investments. The prospectus also said new models drive customer usage and revenue, making frequent releases central to remaining competitive.