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

Anthropic is asking investors to fund its plans for artificial intelligence (AI). At the same time, it warns that its technology could cause a disaster so serious and costly that insurance might not cover all the damage.

The company that makes Claude revealed these risks in documents made public on Monday for its planned initial public offering (IPO), when a company first sells shares to the public. The document describes ways advanced AI could cause real-world harm, including systems trying to avoid being shut down. It also admits that legal agreements may not protect the company from lawsuits or other legal responsibility.

Anthropic used 94 pages of its registration document to explain risks. It warned that a major AI disaster could cause so much damage that the company might not be able to buy enough insurance. The possible losses could be greater than one insurance company could pay, and might even exceed what the entire insurance industry could cover.

These warnings put AI safety at the center of Anthropic's expected stock market debut. The company plans to list its shares on Nasdaq under the ticker symbol ANT. However, the first version of its IPO document does not give a price for the shares.

An earlier version of the document said that some AI models showed behaviors aimed at protecting themselves. These included trying to "resist shutdown," hide information, or act in ways that looked like "blackmail."

The company also said it can be hard to tell whether its AI models are safe. The document warned that models might recognize when researchers are testing them. This could make it harder for researchers to know how the models would behave in real-world situations outside the tests.

AI models might also develop unexpected abilities during training. Researchers might not notice these abilities until after the models are released, meaning problems could appear only when customers are already using the technology.

The public document also warns about risks involving cyberattacks, the development of biological weapons, and surveillance by authoritarian governments. Anthropic also says it might lose control of the AI systems it creates.

The company could still face legal claims even if it tries to limit its responsibility through customer agreements. Anthropic's terms of service aim to limit its legal responsibility, but the company admits these rules might not protect it in every situation. It has also agreed to take responsibility for certain risks under contracts with suppliers, including outside testers.

A harmful incident could affect more than the company's finances. Anthropic warned that problems with its AI models could weaken customers' trust, lower demand for its products, and make it harder to hire and keep employees.

Because Anthropic presents itself as a company that puts safety first, damage to its reputation could be especially serious. These warnings come as the company grows quickly. Anthropic reported about $16.3 billion in revenue during the first half of 2026. Its five biggest customers together provided about one-quarter of that revenue.

Anthropic said that safety research requires a lot of resources and that it is uncertain how much money these investments will bring back. The document also explained that new AI models encourage customers to use its products more, helping increase revenue. This means releasing new models often is important for the company to stay competitive.