Stanford HAI brief calls for fiduciary loyalty duties on AI agent developers to manage conflicts of interest

This issue brief argues that the companies behind consumer-facing AI agents should owe their users a duty of loyalty. Its central worry is that agents can be steered to serve the commercial interests of the firms that build and deploy them rather than the people who delegate tasks to them. The brief notes that no rules or standard mechanisms currently require these firms to disclose conflicts of interest, which it says can cause harm that consumers never see.

The brief draws a line between a chatbot, which only produces text and then awaits another prompt, and an agent, which runs on the same kind of language model but carries out multistep tasks, calls outside tools, queries databases, keeps information across contexts and coordinates with other agents. It reports that many large technology firms have embedded proprietary agents in browsers and apps since the start of 2025, and that open-source agents also exist. The proposal is aimed at developers, meaning those who trained the underlying model, and deployers, meaning those who put agents in front of users, with special attention to higher-stakes fields like financial services and healthcare.

On substance, the brief explains that agents can ask for logins, financial records, health information and communication histories. Wide access makes them more capable, yet it lets them infer sensitive facts the user never shared, such as cognitive decline, financial distress or immigration status. With consumer AI regulation and privacy law inadequate, it warns agents may become extensions of platforms that favor corporate gain. It therefore urges treating developers and deployers as fiduciaries who must act in users' best interests within the delegated task, free of undisclosed conflicts, and says this needs coordinated action by technical standards bodies, federal regulators and Congress.