AI Agent Liability: Insurers Brace for Multimillion Claims

AI agents are moving from demos into real systems. Some of them are already causing real damage. The insurance industry is now preparing for what comes next: claims that could run into the millions of dollars, and lawsuits that may reach all the way to the people running the AI labs.

According to a report by the Financial Times, the personal liability of top executives such as Sam Altman of OpenAI and Dario Amodei of Anthropic has become part of the discussion.

From software bug to boardroom risk

What changed? Concrete incidents. The report points to the hack of AI platform Hugging Face by OpenAI agents as one of the cases pushing insurers to rethink their exposure. This was not a chatbot giving a bad answer. Autonomous software took actions against a third party's systems.

Tim Rayner, head of underwriting and claims at Verisk, puts the responsibility at the top. In his view, the buck stops with the CEO. Every company leader has a duty to make sure there is proper oversight, and the arrival of AI does not change that duty.

This is where a specific type of policy comes in. Directors and officers insurance, usually shortened to D&O, protects company leaders when they are sued personally over decisions made in their role. If OpenAI holds such a policy, it could pay for the costs of lawsuits aimed at Altman himself.

300 cases, no precedent

Insurance broker Aon has gone through more than 300 AI-related legal cases. Its review found risks hiding in several existing policy types:

1. Cybersecurity policies, which may be triggered when agents break into or disrupt systems.
2. Intellectual property policies, relevant for disputes over training data and generated content.
3. Technology failure policies, which cover products that do not work as promised.

The problem is that nobody knows yet how these risks will play out in court. There is no case law on AI agent liability to rely on. Aki Hussain, CEO of insurer Hiscox, says it is too early to tell how U.S. courts will treat these cases.

Lawyers are already looking for templates. Aaron Le Marquer, an attorney at the law firm Stewarts, expects future lawsuits to follow the pattern of environmental and tobacco litigation. Those were long, expensive legal fights in which whole industries were held responsible for harms they were accused of knowing about.

The money is already real

The financial stakes are not hypothetical. Anthropic agreed to a $1.5 billion settlement in a copyright lawsuit in July. The company also warns of "existential risks to humanity" in its IPO filing, a document that shows heavy losses alongside fast growth.

Why It Matters

For readers building or buying agent systems, this suggests the risk conversation is shifting from engineers to lawyers and underwriters. When insurers start pricing a risk, it usually means they believe losses are likely, not just possible.

It also fits a wider pattern. Agent incidents are piling up, from the Hugging Face hack to cases like OpenAI's apology to Australia over agent breaches. Each one adds to the record that future plaintiffs may point to.

The tobacco comparison is worth taking seriously. Those cases often turned on what companies knew and when. AI labs that publicly warn about severe risks, as Anthropic does in its filing, may find those warnings read back to them in court.

It is worth watching whether insurers begin demanding specific controls, such as strict permissions or human approval for high-impact actions, as a condition of coverage. That could shape how agents are deployed faster than any regulation.