Courts do not currently treat an AI system as a legal person that can be sued on its own. Instead, lawsuits over AI mistakes are directed at the company that built, deployed, sold, or relied on the system, using the legal theories that already existed before AI became widespread.

This means a plaintiff harmed by an AI-driven decision generally has to fit the facts into a familiar framework, like showing the company was negligent in how it trained, tested, or deployed the system, or that the system failed to perform as promised under a contract or warranty.

Depending on the relationship between the plaintiff and the company, potential claims can include negligence, product liability, breach of contract, breach of warranty, or violation of a specific consumer-protection or industry regulation.

The right theory depends heavily on facts like whether the AI was sold as a product, offered as a service, embedded in a larger system, and whether the plaintiff was a customer, a bystander, or someone the AI's output was used against.

Because AI systems can behave in ways that are hard to fully explain or predict, courts are still developing how to apply concepts like foreseeability, causation, and reasonable care to machine-learning-driven decisions.

Questions like how much responsibility falls on the company that built the underlying model versus the company that deployed it for a specific use are actively being litigated and have not been resolved uniformly across jurisdictions.

Start by identifying exactly what happened, what the AI system was supposed to do, and what harm resulted, since these facts determine which legal theory fits best.

Preserve any records of the interaction, including outputs, correspondence, and terms of service, and consult with an attorney familiar with the relevant industry, since AI liability cases often depend on technical details specific to the sector involved.