techpolicy.press web signal

Warner's Draft AI AGENT Act Targets Platform Agent Power

TL;DR

  • Senator Mark Warner released a discussion draft of the AI AGENT Act on June 29, 2026, the first federal bill aimed at agentic AI power.
  • The bill would require platforms with at least 50 million U.S. customers to support an interoperable interface for user-chosen 'custodial user agents.'
  • The FTC would register these agents and enforce fiduciary-style duties, though the draft itself flags auditing and agency independence as open problems.

Senator Mark Warner just put the first real marker down on agentic AI, and the interesting thing is not that Congress is finally paying attention, but where the bill chooses to fight. On June 29, 2026, Warner released a discussion draft of the Artificial Intelligence Access, Gatekeeper Exchange, and Nondiscriminatory Transfer Act, the AI AGENT Act, and Rutgers law professor Ellen P. Goodman walks through it in Tech Policy Press. It is not a safety bill, not a training-data bill, not a frontier-model bill. It is an e-commerce interoperability bill dressed for the agent era.

The core move is a platform-access mandate. Large online platforms, defined as those with at least 50 million U.S. customers or subscribers, would have to maintain an interoperable interface that lets users deploy their own custodial user agents, or CUAs. Goodman quotes the draft's definition of a CUA as a software agent 'expressly authorized by a user to interact with a large online platform provider on that user's behalf in a transparent, documented, scope-limited, and revocable manner.' The worked example is Amazon: instead of being funneled through Alexa, a shopper could point their own agent at the store. The FTC would register those agents and enforce the regime under its unfair-or-deceptive-practices authority, with fiduciary-style duties on the CUA side including data safeguarding, real-time recordkeeping, and a bar on unauthorized delegation to other agents.

The reason to take this seriously is the framing Goodman leads with. 'If recommendation algorithms gave online platforms the ability to control what people see online, agentic AI offers them an opportunity to control what people do.' Once an agent is the thing that actually clicks buy, whichever party owns that agent becomes market infrastructure, and Warner's answer is to give users a statutory right to bring their own.

The honest caveat is that the discussion draft flags its own weak spots. Reliable auditing of agent behavior does not yet exist, determining a user's real preferences is hard, and Goodman notes the FTC's independence following Trump v. Slaughter is compromised, which matters a lot for a bill that leans on FTC enforcement. What the article does not give you is any specifics on penalties, private rights of action, or how a platform's in-house agent and a user's CUA would resolve conflicts in practice.

Still, this is the first draft that treats the agent layer as the choke point worth regulating, and any consumer-agent startup that has been quietly assuming the big platforms would eventually wall them off should read the CUA definition line by line.

Shared on Bluesky by 2 AI experts