AI agents promise to "run the business," but who is liable if things go wrong?

The pitch — hands-off management, 24/7

It has been reported that vendors are pitching autonomous AI agents as the next frontier of automation: hire them, hook them into your systems, and they will handle tasks from scheduling and customer support to procurement and even contract negotiations. Sounds like magic. Sounds dangerous. For companies chasing efficiency, the lure of an always-on digital executive is irresistible. But when an agent makes a mistake — signs the wrong contract, orders the wrong goods, or triggers an erroneous transfer — who pays the bill?

The legal fog

Allegedly, some providers are already trying to paper over the risk by putting broad liability limits and indemnities into their contracts, pushing responsibility onto customers. That raises old legal questions in a new costume. Is this product liability? An agency problem? Negligence? Or a novel third category that courts will have to invent? There are practical fixes being discussed — mandatory audit logs, human-in-the-loop requirements, insurance products — but they don't solve the deeper question of moral and legal accountability when decision-making is outsourced to code.

Regulators, insurers and the courtroom

Regulators around the world are paying attention. It has been reported that policymakers and standards bodies are debating whether existing regimes — consumer protection, corporate law, product-safety rules — are up to the job or whether bespoke rules are needed for autonomous agents. Insurers smell opportunity; lawyers smell test cases. Expect a flurry of litigation, a scramble for precedent, and a patchwork of policies before anything resembling clarity emerges. Think of it as an industrial revolution in slow motion, where the factory is software and the injuries are contractual and financial.

Who picks up the tab?

There’s a human moment in all this: companies imagine a future where software quietly does the heavy lifting, then wake up to a multi-million-dollar IOU. Who will be left holding the bill — the engineers who built the agent, the vendors who marketed it, the executives who bought it, or the customers who relied on it? The answer will shape how quickly businesses adopt these systems. For now, the promise is dazzling; the risks, very real. Regulation, insurance and a few court rulings will decide whether AI agents become trusted colleagues or costly gambles.

Sources: The Register