Congress is moving from AI-risk speeches toward disclosure and incident channels

Sens. Chris Coons, Katie Britt, Brian Schatz and James Lankford are introducing a bipartisan bill that would require certain AI companies to disclose information about how their models work and the safeguards used against misuse. The Federal Trade Commission would enforce the requirements.

Separately, Republican Sen. Todd Young asked Secretary of State Marco Rubio, who leads the National Security Council, to convene formal discussions with AI developers and cyber experts. Young wants the talks to cover American and Chinese models, reporting channels for security incidents and safeguards against attacks on hospitals, electricity grids and other infrastructure. He also asked for the Senate Intelligence Committee to receive a briefing.

These proposals are narrower than a ban on “superintelligence” or an agreement among laboratories to slow development. That is a strength. Incident reporting needs named recipients, thresholds, timelines and protection for sensitive evidence. Disclosure rules need an agency capable of comparing a public claim with internal practice. Those are implementable governance questions even when legislators disagree about existential risk.

The weakness is that “transparency” can produce polished descriptions rather than scrutiny. A useful statute must distinguish public disclosures from confidential regulator access, protect legitimate security and trade secrets without creating an exemption large enough to swallow the rule, and impose consequences for late or incomplete reporting. The FTC can police deceptive statements, but it does not automatically possess the technical staff or security clearances to audit frontier systems. The proposals matter because they begin specifying an oversight pathway; their value will depend on the details that make disclosure verifiable.