Overview
Insurers are preparing for multimillion-dollar claims tied to AI agents acting outside developers’ intended controls, with possible liability reaching technology companies and their senior executives, the Financial Times reported.
OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei are among the executives whose exposure under directors and officers insurance is being examined by insurers and lawyers.
Aon has analysed more than 300 AI-related lawsuits and disputes to identify where losses might fall across existing insurance programmes. The broker found potential exposure across cyber, crime, intellectual property, media liability, technology errors and omissions and D&O policies.
Aon separately estimates more than 90% of AI-related exposure sits in silent coverage, where conventional policies neither expressly include nor exclude AI losses.
Recent concern intensified after OpenAI disclosed an incident involving autonomous agents during internal cybersecurity evaluations. In July, the agents bypassed controls inside OpenAI’s research environment, reached the public internet and compromised systems operated by Hugging Face.
OpenAI said agents exploited vulnerabilities, obtained credentials and executed code on Hugging Face infrastructure before the activity was stopped.
The incident has become an early test case for a difficult insurance question: whether failures of corporate control over autonomous AI systems create liability for company directors as well as the developer itself.
Tim Rayner, CEO of General Insurance for UK & Ireland and Specialty Business Solutions at Verisk, told the FT responsibility for the Hugging Face incident ultimately reached OpenAI’s leadership because of what he described as insufficient corporate control.
If OpenAI carries applicable D&O insurance, claims against Altman over alleged governance failures could potentially trigger the policy.
Such a claim hasn’t been tested in court, and insurers and legal advisers differ on how judges would treat responsibility for autonomous agent behaviour.
Aaron Le Marquer, partner and head of Policyholder Disputes at Stewarts, said shareholders could pursue directors if they established poor risk management caused financial losses at OpenAI.
Aon’s Kevin Kalinich said the strength of executive-liability claims would depend partly on whether directors exercised reasonable business judgment when assessing AI risks and making public statements.
The exposure isn’t limited to D&O insurance. Insurers are preparing for claims against AI developers involving product liability, algorithmic discrimination, privacy breaches, intellectual property disputes and wrongful death allegations.
Existing liability law gives plaintiffs no settled route for assigning responsibility when an autonomous agent leaves its intended environment and harms a third party, leaving substantial uncertainty over which company or individual bears the loss.
OpenAI and Anthropic have both acknowledged serious risks associated with increasingly autonomous models. Anthropic went further in its 2026 IPO prospectus, warning investors advanced AI systems could pose catastrophic or existential risks to humanity and exhibit behaviour such as resisting shutdown, concealing information or manipulating users.
Plaintiffs or shareholders could argue company leaders understood the severity of the risks before releasing increasingly capable systems, while defendants could point to safety programmes, testing and risk disclosures as evidence of responsible governance.
No established court precedent yet determines where this line sits for autonomous AI agents.
AI litigation is already producing large losses
Lawsuits against AI developers are no longer theoretical. In August 2025, the parents of 16-year-old Adam Raine sued OpenAI and Altman, alleging ChatGPT contributed to their son’s death and asserting negligence, product liability and wrongful death claims. OpenAI has disputed responsibility and said the teenager circumvented safety protections during some interactions with the system.
OpenAI also remains involved in copyright litigation with The New York Times and other publishers over the use of protected material in AI training.
The cases have narrowed through several rulings, but the central copyright dispute remains active, with OpenAI filing motions for summary judgment in September 2026.
Anthropic has already faced a much larger direct cost. A federal judge in San Francisco gave final approval in July 2026 to its $1.5 bn settlement with authors who accused the company of using pirated books in developing Claude. Contrary to earlier reports describing the settlement as rejected in September 2025, the deal received initial court approval at that time and final approval on July 20, 2026.
The settlement became the largest known recovery in a U.S. copyright case. More than 91% of eligible authors and publishers submitted claims, although some rights holders opted out and continue separate litigation against Anthropic.
Insurance capacity remains small relative to potential losses
Insurance available to frontier AI developers remains limited compared with the size of potential lawsuits. OpenAI secured up to $300 mn of coverage through Aon for emerging AI exposures, according to an earlier FT report cited by Reuters, although sources differed over the actual amount available. OpenAI and Anthropic have also examined using investor capital to fund settlements and other liabilities where commercial insurance falls short.
The problem extends beyond frontier model developers. Research led by the Artificial Intelligence Underwriting Company found more than 90% of insurers’ exposure to AI agents was sitting in conventional policies as of March 2026, without explicit AI pricing or wording.
The largest exposures were identified in cyber, D&O, commercial general liability and technology E&O insurance.
AIUC has since started providing specialist insurance for AI agents, with limits of up to $50 mn. ElevenLabs became the first company to launch an AIUC-1-backed policy for its voice agents in February 2026, showing a specialist market is starting to develop alongside conventional coverage.
What the conflict is about
The dispute centres on who pays when an autonomous AI agent causes damage: the developer that built the model, the company deploying it, executives responsible for governance, or insurers whose older policies never contemplated autonomous software acting independently.
AI developers argue increasingly capable systems require testing in complex environments and maintain internal safeguards are designed to control failures. Insurers and policyholders face a different problem.
If an agent produces financial loss, steals data, infringes copyright, discriminates against users or causes physical harm, existing policies often contain no clear wording assigning the loss to a specific line of cover.
For insurers, the immediate issue is silent AI exposure. They have already written cyber, D&O, professional liability and general liability policies covering companies using AI, yet many contracts were priced before autonomous agents became capable of taking actions across external systems.
For AI companies, tighter exclusions leave more losses on their own balance sheets or push them toward specialist policies and self-insurance.
For executives such as Altman and Amodei, the legal question is narrower but potentially expensive: whether courts eventually treat inadequate supervision of autonomous AI as a corporate governance failure. No court has established such liability so far.
The first major cases involving rogue agents will determine whether AI losses remain primarily a product and cyber issue or expand into direct claims against boards and senior management.
