Warner's AI AGENT Act Would Force Platforms Open to Agents

The first federal attempt to regulate agentic AI has arrived as a discussion draft, and the details are worth reading closely. Senator Mark Warner, a Democrat from Virginia, released the AI AGENT Act on June 29, 2026, and an analysis in Tech Policy Press by Rutgers Law professor Ellen P. Goodman lays out what it does and where it strains against what today's agents can actually do.

The bill borrows a familiar shape from Warner's earlier work on platform interoperability. Large online platforms, defined as those with 50 million or more U.S. customers or subscribers, would have to maintain portals that let independent 'Custodial User Agents,' or CUAs, act on a user's behalf across shopping, content selection and account management. The CUAs themselves would owe fiduciary duties: they must 'not act to benefit the CUA at the user's expense, in ways that cause reasonably foreseeable harm,' must safeguard privacy, keep real-time action records, and get express permission before delegating authority. The FTC would register them, with recognized independent certification entities able to create a rebuttable presumption of compliance.

Why this matters is the plumbing more than the politics. If your agent has to go through Amazon's checkout, Apple's app store or Google's search box on the terms those platforms set, 'agentic AI' mostly means whatever the incumbents let it mean. Warner is trying to pry that open before the market settles, borrowing the same interoperability instinct he brought to social media portability in his 2019 ACCESS Act.

The honest caveat, and Goodman is direct about it, is that the technology to enforce the bill does not exist yet. The draft calls for 'new technical standards for agentic operation and verification: reliable records of what an agent did and why; a kill-switch to allow a user to revoke consent from an agent; and security and privacy controls better than the state of the art.' Reliable chain-of-thought explanations, she notes, are 'not available now,' and agents 'may act over an extended period of time, without their behavior having been specified in advance,' with errors compounding as the process unfolds. What the piece does not resolve is how a politicized FTC would actually run a registration regime this technical, what happens with multi-agent stacks, or who anoints the independent certifiers.

Still, the direction is the interesting part. If the interoperability duty holds in some form, third-party agent builders get a foothold against the platform-owned assistants, and users get a plausible path to 'bypass dark patterns, find the best deals, and fact-check at scale' without living inside one company's stack.