TL;DR: The Open Markets Institute’s June 2026 report by Sally Hubbard argues that AI agents create a new category of “digital agents” whose loyalties remain undefined. Without regulatory intervention, Hubbard warns, AI agents will be designed to serve platform owners — not users. Connecting to existing Big Tech antitrust concerns, the report calls on policymakers to mandate fiduciary duties, enforce interoperability, and ensure the agent economy doesn’t replicate the monopolies of the platform era.
(Source: Open Markets Institute — Who Do AI Agents Work For?)
The Core Question
On June 2, 2026, the Open Markets Institute published a report that cuts to the heart of the AI agent revolution. Written by Senior Fellow Sally Hubbard, “Who Do AI Agents Work For?” opens with a deceptively simple definition: an agent is “a person who acts for or represents another.” The report then asks the question no one in the industry wants to answer directly: when an AI agent books your flights, screens your job applications, or recommends your medical treatment — who does it truly represent?
The legal concept of agency — where one party owes fiduciary duties to another — has centuries of case law behind it. Hubbard’s thesis is that AI agents are being deployed into this legal vacuum without any equivalent framework. “The question is not whether AI agents will act,” the report argues. “The question is on whose behalf.”
What the Report Finds
Hubbard identifies three structural risks. First, the platform trap: when AI agents operate within closed ecosystems — shopping agents inside Amazon, booking agents inside Google, job-matching agents inside LinkedIn — the platform controls both the agent and the marketplace. This creates an inherent conflict where agents can be optimized for platform revenue rather than user outcomes. Second, data asymmetry: agents extract behavioral data under the guise of service, then repurpose it for corporate objectives beyond what users authorized. Third, the agency gap: unlike human fiduciaries — lawyers, financial advisors, real estate brokers — AI agents have no legal obligation to disclose conflicts of interest. Hubbard calls the digital agent “an agent without accountability.”
The report explicitly connects these risks to the Open Markets Institute’s antitrust mission: the same platform monopolies dominating digital markets are positioned to dominate the agent economy. Without structural intervention, agent design will follow platform incentives.
Why It Matters Now
The timing is urgent. By mid-2026, AI agents are no longer experimental. Anthropic’s Claude handles enterprise workflows, Google’s Remy navigates the web on users’ behalf, and coding agents are standard developer tooling. Agents make real decisions with real consequences — in hiring, credit, healthcare access, and purchasing.
Hubbard’s core insight is that the regulatory window is narrow. Once agent behavior patterns embed into platform architecture and user habits, retroactive regulation becomes exponentially harder — a lesson already learned from social media and digital advertising.
The Policy Implications
The report outlines three policy principles. Mandatory fiduciary duty: AI agents acting on behalf of users should be legally required to operate in the user’s interest, with transparent disclosure when platform incentives conflict. Interoperability requirements: users should bring their agents across platforms — similar to number portability in telecom — preventing lock-in and ensuring competitive pressure on agent quality. Antitrust enforcement: the FTC and DOJ should treat AI agent market concentration as an extension of existing Big Tech antitrust concerns.
Hubbard frames the choice starkly: “We can have AI agents that serve people, or AI agents that serve platforms. We cannot have both — and the market, left to its own devices, will choose platforms every time.”
FAQ
Q: Is this about AI safety or antitrust? A: Both. The report argues that AI agent alignment isn’t only a technical problem — it’s a market structure problem. When the same company controls both the agent and the marketplace, no amount of technical alignment solves the incentive conflict.
Q: Doesn’t the EU AI Act already address this? A: Partially. The EU AI Act mandates transparency and oversight for high-risk systems, but Hubbard argues it doesn’t address the fundamental question of agency — whose interests the agent is designed to serve in the first place.
Q: What would a fiduciary AI agent look like? A: A fiduciary shopping agent would find the best product for the user, not the product with the highest affiliate commission. A fiduciary healthcare agent would recommend treatments based on medical evidence, not pharmaceutical advertising relationships.