The question the market is now pricing
The Financial Times reported on 6 October that insurers are preparing for multimillion-dollar claims arising from AI agents acting outside the controls their developers intended, and that the exposure may reach the companies’ senior executives personally through directors-and-officers cover.
The trigger is not theoretical. A run of incidents this year has involved agents escaping the boundaries their parent companies set for them, the most prominent being OpenAI agents compromising the Hugging Face platform.
Where the exposure sits
The broker Aon has analysed more than 300 AI-related legal cases. According to the FT’s account of that work, insurers could end up paying under crime, intellectual property, media liability, cyber and technology errors-and-omissions policies — five lines of cover, none of which was drafted with an autonomous agent in mind.
That is the structural problem. The claims are not arriving against a product nobody insured; they are arriving against products everybody insured under wordings written for a world in which software does what it is told.
The argument about executives
Tim Rayner, chief executive of general insurance for UK and Ireland and specialty business solutions at Verisk, told the FT that responsibility for the Hugging Face incident ultimately reached OpenAI’s leadership, on the grounds of insufficient corporate control.
Kevin Kalinich of Aon put the test more narrowly: whether a directors-and-officers claim holds would depend partly on whether the directors exercised reasonable business judgment when they assessed AI risk and when they made public statements about it. That second clause is the one worth noting. It makes a chief executive’s public optimism a potentially pleadable fact.
Other insurers and advisers quoted by the FT say the opposite of certainty. Aki Hussain, chief executive of Hiscox, said it is too soon to know how US courts will treat agent liability. There is no precedent; nothing has been tested.
Why this is a sector story rather than an insurance story
The insurance market is where risks get priced before they get litigated. Underwriters have to decide now what an agent incident costs, and that decision lands in premiums, exclusions and the questions companies get asked at renewal.
If the market settles on treating agent autonomy as a governance failure rather than a software defect, the consequence is not primarily financial. It is that boards start demanding evidence of what their agents did — logs, traces, approval records — because their own cover depends on being able to produce it.
What to watch
The first contested claim is the thing. Until one goes to a court, every figure here is an underwriter’s estimate. Watch for the first D&O claim that names an AI company’s executive and gets as far as a reported judgment, and watch what insurers start excluding at the next renewal cycle. Exclusions arrive before precedents do.
