What the bill does
Four senators have introduced a bill that would require artificial intelligence companies to publish what their systems collect and how they are guarded, and would give the Federal Trade Commission the job of enforcing it.
S.5471 was introduced on 23 September by Chris Coons, Democrat of Delaware, with Katie Britt, Republican of Alabama, Brian Schatz, Democrat of Hawaii, and James Lankford, Republican of Oklahoma, as original cosponsors. Its official title is a bill “to require disclosure of certain information relating to artificial intelligence systems”, and it is being called the AI Systems Transparency Act. It was read twice and referred to the Senate Committee on Commerce, Science, and Transportation.
Semafor reported the bill on Thursday. According to that account, covered companies would have to disclose what data their models collect, what safeguards protect children and adults, and what measures are meant to stop systems being misused or operating dangerously.
Two audiences for one disclosure
The structure is the interesting part. Rather than a single public statement, the bill as described would require separate disclosures for consumers and for independent researchers and evaluators — an admission that the document a shopper needs and the document a safety auditor needs are not the same document.
The safeguards named cover child safety, mental health, privacy, cybersecurity, and the risk of systems exceeding human control. Companies would also have to describe their models’ structure, their own policies and the policy violations they see most often, and to update all of it when they ship a new model or substantially change an existing one.
It applies to closed and open-source models alike, with criteria determining which companies fall under the FTC’s oversight.
Where it comes from
This is not the first move by these four. In December 2025 the same senators wrote to eight AI companies asking for more information about model capabilities and user risk. The bill is what follows when letters do not produce a standard.
“Trust has to be earned, and right now families are being asked to trust AI companies with very little information,” Lankford said. Coons framed it as a right to know what companies are actually doing to put guardrails on large models.
How it compares
Placed next to the other AI bills in this Congress — a proposed permanent ban on artificial superintelligence, a new federal agency, a framework for AI-enabled industries — this one is deliberately modest. It does not tell anyone what to build or what to stop building. It tells them to write down what they already do.
That is also what makes it plausible. A referral to Commerce is the beginning of a process, not the end of one, and most bills die there. The signal to watch is whether the committee schedules a hearing, and whether any of the companies that received the 2025 letters say in public what they think of being asked the same questions with the FTC behind them.
