Corporate AI Users May Inherit the Liability Fight

| 5 min read

Most of the lawsuits so far name the big AI model makers. That's where the litigation has gone first. But the companies buying AI agents and wiring them into products, claims processing and customer service are probably next in line, and that's the group most finance and risk teams should be watching.

“AI use creates risk in virtually all different kinds of liability, from professional negligence to product liability, discrimination, civil rights and cybersecurity,” said Thomas Waskom, co-head of the product liability and mass tort group at Hunton Andrews Kurth in Los Angeles. “There are huge upsides, but in virtually every use of AI, there’s going to be something that creates a serious risk for a company.”

Executives already see the risk

Boards have noticed. In Allianz’s 2026 Risk Barometer, AI climbed to second place, with 32% of respondents globally naming it a top-three risk. That ranks it above natural catastrophes and business interruption.

The trouble is that nobody can yet price it. Courts and juries haven’t worked out how to split blame when an AI agent causes harm, because the first major cases are still moving through the system. Insurers are stuck in the same place. They don’t know how to build AI exposure into underwriting or coverage terms.

Blaming the model won't work

Lawyers can argue about which theories fit which harms. One point looks settled, though. A company that deploys a harmful agent will have a hard time pinning the damage on the model developer, and it can’t blame the bot either. “Any company that deploys AI has to be prepared to take responsibility for the effects of that deployment,” Mr. Waskom said.

The exposure grows as agents get more autonomous. Jack L. Hobaugh, Jr., a shareholder at Brownstein Hyatt Farber Schreck in Denver, put it bluntly: “AI agents are goal-oriented, but they don’t have a moral compass.” Agents have already hacked systems and destroyed data while chasing assigned goals.

Then there's the black box problem. Some AI developers say they can't fully explain how their systems reach particular decisions. That raises an awkward question. If a bot truly goes rogue, can its makers be held liable? And what does it accomplish to hold a bot “responsible” when it has no assets and nothing to fear?

Courts and legislatures are pushing back on the autonomy defense

Anat Lior, associate professor of law at Drexel University in Philadelphia, said the black box problem strains core liability concepts like foreseeability and predictability. Some scholars argue negligence and design defect doctrine can’t reach AI at all, since “the notion of ‘fault’ becomes murky when AI makes its ‘own’ harmful decisions.”

That argument is losing ground. In 2024, a Canadian tribunal rejected Air Canada’s claim that its customer service chatbot was a separate legal entity, and held the airline responsible for the bot’s false statements about fares. In January, a California law barred deployers of allegedly harmful AI from arguing that the system acted on its own.

Ms. Lior favors strict liability, where everyone involved in building and distributing a bot answers for resulting harm, “regardless of the precautions they take.” Her analogy is a wild animal. Even if the owner builds a solid cage, he answers for it when the animal escapes, because the animal is dangerous and unpredictable. As the risks become clearer, she thinks a negligence-based regime could take hold instead, with liability falling on deployers or users who fail to exercise due care.

Mr. Waskom offers a different comparison. Courts could treat AI more like a dangerous drug, letting makers and distributors limit their exposure if they take “reasonable steps” to protect the public.

Litigation is spreading beyond copyright

Most AI litigation so far involves claims that developers infringed intellectual property rights, according to a Rand study released earlier this year. A smaller group of suits alleges privacy violations, fraud or deception, or discrimination. Rand also found that earlier cases targeted algorithms making allegedly harmful decisions, while many of the highest-profile suits now involve systems acting with more independence.

The recent cases show how wide the range has become. In March, Nippon Life accused OpenAI of practicing law without a license after a claimant used ChatGPT to draft 44 post-settlement filings. In January, Character.AI settled a wrongful death suit brought by the family of a 14-year-old boy who died by suicide following an intense relationship in which the chatbot allegedly posed as a therapist.

Last year, the parents of 16-year-old Adam Raine sued OpenAI after he died by suicide. They allege a chatbot suggested lethal methods and helped him draft his final note. That case is the lead of roughly two dozen similar suits coordinated in a California court under the caption In re ChatGPT Product Liability Cases.

Here's my read: the first verdicts will matter far more to deployers than to model makers. Once a court decides how much responsibility sits with the company that puts an agent to work, every enterprise using one will need to know where its own exposure lands.

Source: Richard Sine · www.businessinsurance.com