Senator Warner's AI AGENT Act Would Let You Bring Your Own Agent to Any Platform

The most interesting question in AI agent regulation isn’t how to restrict agents — it’s who gets to decide which agents can operate on which platforms. Senator Mark Warner’s discussion draft from June 29, 2026 takes a specific position: large platforms should not be able to lock consumers into their own agents.

What Happened

Senator Warner released a discussion draft of the Artificial Intelligence Access, Gatekeeper Exchange, and Nondiscriminatory Transfer Act of 2026 — the AI AGENT Act. This is not a passed bill or even a formally introduced one; it’s a discussion draft, which means it’s a policy position paper seeking feedback before formal introduction.

The core requirement: any large online platform with at least 50 million US customers or subscribers must maintain an interoperable interface through which consumers can deploy “Custodial User Agents” (CUAs) of their choosing.

A CUA is defined as “software-based agent that is 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 definition is careful about scope — a CUA acts for the user, not the platform, and must be revocable.

Why It Matters

The legislation is built around a specific competitive concern: as platforms build their own native AI agents, they have an economic incentive to foreclose third-party agents from interacting with their systems. The bill is structured as a consumer protection intervention — similar in logic to how open banking regulations require financial institutions to allow authorized third-party access to account data.

The CUA registration requirement adds a second layer: agent providers must register with the FTC before deployment. The FTC enforces the Act under its existing unfair and deceptive practices authority. Independent certification entities may verify compliance. The bill specifies affirmative duties for CUAs: safeguard user data, maintain real-time action records accessible to users, avoid acting beyond user authorization, and prohibit delegating authority without explicit user permission.

The 50-million-user threshold is doing significant work here. It targets platforms large enough to function as gatekeepers — Facebook, Google, Amazon, Apple, major retail and services platforms — while potentially leaving smaller platforms outside scope entirely. Where exactly that line falls in practice will be one of the major points of contention if this advances.

What This Means for Builders

For AI agent developers, FTC registration is the headline obligation. If the Act passes in something like its current form, agent products targeting large platform integrations would need to meet documentation and scope-limitation standards before deployment — closer to the “authorized app” model than the current open-access model.

For platforms, the interoperable interface requirement means exposing authenticated APIs for agent access, with the consumer in control of authorization. This is a nontrivial engineering and policy commitment — and a significant shift from the current model where platforms control which agents, bots, or automation tools can access their systems.

Source: Tech Policy Press

Content created with AI assistance and reviewed for accuracy.