The current regulatory focus on transparency is proving insufficient to address the structural risks posed by AI agents embedded in our digital infrastructure. While disclosure requirements aim to inform users, they do not prevent the underlying problem of deceptive steering, where AI agents are designed to prioritize the commercial interests of their developers over the needs of the user. A new proposal from Stanford HAI, titled Designing Loyalty: AI Agents and Conflicts of Interest, argues that we must move beyond mere transparency and impose a formal fiduciary duty on those who build and deploy these systems.
Published on August 25, 2026, this brief marks the first time a major academic institution has formally advocated for classifying AI developers and deployers as fiduciaries. This shift represents a fundamental change in how we approach AI governance, moving from a model of passive disclosure to one of active, legally binding loyalty. By imposing a duty of loyalty, the proposal would require entities to act in the best interests of their users within the scope of any delegated task, effectively prohibiting the use of undisclosed conflicts of interest to manipulate user outcomes.
Since early 2025, major technology companies including Amazon, Google, Anthropic, OpenAI, Perplexity, Meta, and Microsoft have integrated proprietary AI agents directly into browsers and applications. These agents now act as intermediaries for high-stakes decisions in finance and healthcare, creating a significant power imbalance. When an agent is designed to serve both the user and the developer, the potential for conflict is inherent. The Stanford HAI proposal addresses this by suggesting that in these high-stakes domains, the agent must be legally bound to the user’s interests first.
This academic proposal arrives at a critical juncture, complementing existing enforcement actions from federal regulators. The Federal Trade Commission (FTC) issued a proposed policy on July 1, 2026, specifically targeting AI-driven deceptive steering under Section 5 of the FTC Act. Similarly, the Securities and Exchange Commission (SEC) has made AI-related disclosures and conflicts of interest a central pillar of its 2026 Examination Priorities. These actions build upon a history of regulatory scrutiny, including the SEC’s March 2024 settlements with Delphia and Global Predictions regarding AI washing, and the agency’s December 2025 Marketing Rule risk alert.
If adopted, the fiduciary framework would necessitate significant changes in how AI agents are designed and managed. Developers would be required to identify, manage, and explicitly disclose any conflicts of interest that could influence an agent’s recommendations. This is not merely a technical challenge but a structural one. It requires a move away from business models that rely on steering users toward preferred products or services, forcing a re-evaluation of how AI agents are incentivized. The implementation of such a policy, however, faces substantial hurdles. Defining the scope of a ‘fiduciary’ relationship in the context of software requires precise legal and technical definitions. There is also the risk that overly rigid requirements could stifle innovation or create compliance burdens that only the largest firms can manage. The Stanford HAI brief suggests a domain-limited approach, starting with healthcare and finance, which may offer a more manageable path forward than a blanket regulation.
Beyond the legal classification, the proposal calls for a suite of supporting measures, including digital agent identifiers, federal privacy legislation, and mandatory reporting for adverse incidents involving AI agents. These recommendations aim to create a comprehensive ecosystem of accountability. By requiring coordinated regulatory action, the proposal seeks to ensure that the duty of loyalty is not just a theoretical concept but a practical standard for the industry.
The structural significance of this proposal lies in its focus on the ‘who’ and the ‘why’ of AI decision-making. By asking whose interests an AI agent serves, policymakers are beginning to address the core tension of the digital age: the conflict between the efficiency of automated agents and the autonomy of the human user. If the agent is a fiduciary, the user is no longer just a consumer of a service, but a principal to whom the agent owes a duty of care.
This shift toward fiduciary duty also highlights the limitations of current ‘AI washing’ enforcement. While the SEC has successfully targeted misleading claims about AI capabilities, those actions do not necessarily address the underlying design choices that lead to biased or self-serving agent behavior. A fiduciary standard would provide a more robust framework for evaluating whether an agent’s design is fundamentally aligned with the user’s best interests.
As the regulatory landscape continues to evolve, the conversation is clearly moving toward more stringent oversight. The alignment between academic research and federal enforcement suggests that the era of self-regulation for AI agents is coming to a close. Whether through the FTC’s focus on deceptive steering or the SEC’s scrutiny of financial disclosures, the message to developers is becoming increasingly clear: the design of AI agents must be transparent, accountable, and, above all, loyal to the user.
Ultimately, the proposal from Stanford HAI provides a roadmap for a more responsible AI future. By grounding the governance of AI agents in the established legal principles of fiduciary duty, it offers a path to mitigate the risks of manipulation and conflict. As these agents become more deeply embedded in our daily lives, the need for such a standard will only grow, making the debate over loyalty a defining issue for the next phase of AI policy.