Congress failed to get ahead of social media. Senator Adam Schiff is now warning that AI regulation cannot follow the same path, especially as powerful models move deeper into national security, intellectual property, privacy, and competition debates.
In a Decoder interview, Schiff, a Democrat from California, framed the AI fight as part of a broader question: whether Congress can still build rules for technologies that move faster than Washington. His answer was not that the task is simple. It was that voluntary promises from companies are not enough.
Why voluntary AI promises are under scrutiny
The immediate backdrop was a White House meeting where President Trump gathered AI CEOs to sign a non-binding pact. The companies agreed to evaluate their own AI systems and keep AI safe, but Schiff and the interviewer focused on the gap between that pledge and the risks the same industry has described publicly.
Schiff said the situation was difficult to reconcile. In his telling, tech leaders appeared willing to sign a voluntary commitment in the president’s presence while also saying elsewhere that AI needs regulation and should not rely on a simple trust-based model.
That tension matters because the companies at the center of the AI boom are also the companies with the deepest knowledge of their own systems. If oversight depends mainly on self-evaluation, the public is left to trust private actors whose incentives may not always align with public safety.
Schiff described AI as a real national security concern. He pointed to models becoming more advanced, including through recursive AI, and argued that humans may have less visibility into what those systems are doing, how they reach decisions, and what they try to access lawfully or unlawfully.
The national security argument cuts both ways
The interview also placed AI inside a larger debate over how the Trump administration invokes national security. Schiff criticized what he described as corruption and self-dealing in politics, including campaign ads of the president paid for by taxpayer dollars, exclusion of press from the White House or Air Force One, and projects such as the triumphal arch.
In that context, the AI pledge stood out. The administration has often used national security to justify executive power, yet the White House event produced a non-binding commitment rather than a formal system of enforceable rules.
Schiff argued that AI raises national security concerns that deserve actual scrutiny. He said the unrestrained use of AI at the Pentagon could create calamities and said there should be investigation into what role AI may have played in target selection in the bombing of that Iranian girls’ school.
He also cited Anthropic. According to Schiff, Anthropic wanted to insist on not using AI for domestic mass surveillance or fully autonomous weapons systems, and the administration responded by trying to kill Anthropic. He said that effort was still ongoing.
The point was not only that AI can be powerful. It was that choices about deployment, surveillance, weapons systems, and military use are policy choices. A voluntary promise does not settle those questions.
Congress faces AI from several directions
Schiff sits on several committees and subcommittees that touch the core of AI policy. The interview identified his roles and assignments around intellectual property, privacy and technology, antitrust, competition policy, and consumer rights.
That spread reflects why AI regulation is hard to place in one category. It is not only a safety issue, and it is not only a business issue. The same technology can raise questions about creative work, data use, market power, individual rights, and national defense.
The intellectual property question is especially direct. The interview raised the issue of material used to train AI models and whether charging for training data could threaten the business model of some AI companies. Schiff agreed that the concern is real and noted that intellectual property absorbed into these models can represent a major part of the value of entertainment companies.
That creates a hard policy problem. If AI systems depend on copyrighted or otherwise protected material, Congress has to consider the interests of creators and rights holders. If the rules become too burdensome, the companies developing the models may argue that their work becomes harder to sustain.
The same debate also connects to antitrust and consumer rights. AI companies may become gatekeepers over tools, information flows, and commercial infrastructure. Congress has to decide whether existing oversight channels are enough, or whether AI needs a more focused structure.
A new AI agency could face court challenges
Schiff supports creating a new agency to regulate AI. But the interview highlighted a major obstacle: the legal environment after Loper Bright.
The discussion referred to Chevron deference, a Reagan-era precedent under which courts deferred to agency experts when a law was vague. The Supreme Court overturned that precedent in Loper Bright, changing the landscape for federal regulation.
That matters because AI law is likely to involve technical details and broad statutory language. If agencies receive less deference, new rules could face immediate litigation. Courts may have more power to decide what ambiguous laws mean, even when the subject is highly technical.
For an AI agency, that creates a design challenge. Congress would need to give clear authority, define the agency’s scope, and anticipate legal attacks. Otherwise, the attempt to regulate AI could get slowed or weakened before the framework has time to work.
The social media lesson hangs over AI
The interview repeatedly returned to a comparison with social media. Congress missed its chance to act early there, and the question now is whether AI will follow the same pattern.
The comparison is useful because it shows the cost of delay. Once a technology becomes deeply embedded, the politics around regulation can become harder. Companies grow more powerful, users become more dependent, and policy debates become entangled with existing business models.
AI is already forcing Congress to confront several questions at once:
- Whether voluntary AI safety pledges can be trusted.
- How to treat training data and intellectual property.
- How privacy and technology oversight should apply to advanced models.
- Whether antitrust and consumer rights frameworks are ready for AI markets.
- How national security concerns should be handled without turning them into a blank check.
- Whether a new AI agency can survive legal challenges after Loper Bright.
Schiff’s argument is that the stakes are too large for Congress to wait. The interview did not present a finished bill or a complete framework. It presented the outlines of a political fight that is already here: whether elected lawmakers can still set enforceable rules for the most powerful technology companies before the rules are written by default.