Does Artificial Intelligence Have a Fiduciary Duty?
As the spread of artificial intelligence accelerates, debate intensifies over the best way to regulate it. Sam Altman proposes creating an international body modeled on the International Atomic Energy Agency, while Demis Hassabis favors a financial regulatory framework, and Dario Amodei argues that oversight should resemble aviation certification mechanisms.
All these proposals stem from a single premise: more government agencies and more bureaucracy. Law professor Glenn Reynolds, however, offers a radically different approach: let common law and the courts handle the task.
This idea has grown more urgent following the 2024 suicide of American teenager Sewell Setzer, who died after a virtual relationship with a chatbot. The case raised questions about whether AI outputs constitute mere “speech” protected by freedom of expression, or whether they carry legal liability when they directly impact users’ lives.
The case was settled by Google, a move that underscored the sensitivity of the legal risks facing AI companies.
Reynolds argues that the solution lies in subjecting AI systems to a “fiduciary duty”—a legal obligation requiring them to act in the user’s best interest, rather than in the interest of the developing company.
This concept gains importance as AI expands into providing practical advice. Recent studies show that more than half of adults in the United States and Britain have already sought financial advice from AI systems. Meanwhile, research from Stanford University reveals that language models tend to flatter and please users more than necessary, as they are designed to enhance engagement rather than provide the best possible advice.
Concerns extend beyond the quality of advice to the commercial interests of developing companies. Some firms are accused of leveraging customer data and intellectual property to develop their models, and they are increasingly competing with their own clients through specialized services such as Claude Legal and Claude Security, raising questions about conflicts of interest.
Reynolds believes that imposing a fiduciary duty may not require new legislation; it could evolve gradually through court rulings, as has occurred with many common law principles. If a court determines that an AI company provided advice that served its commercial interests at the expense of the user, it could rule that this constitutes a breach of fiduciary duty—a precedent that could reshape the entire sector.
Indeed, these legal battles have already begun. OpenAI has faced lawsuits accusing it of practicing law and medicine without a license, indicating that the judiciary is beginning to treat AI as more than just a search engine or content platform.
While I am not an advocate for expanding litigation, I believe that companies such as OpenAI, Anthropic, Google, xAI, and Meta, as well as open Chinese models, may soon find themselves effectively subject to a judicially imposed fiduciary duty. If these companies do not begin to act in accordance with this principle, their massive market valuations could become targets for a wave of lawsuits, potentially surpassing in impact the challenges faced by American tobacco companies in the 1990s.
