- Lead. A bipartisan AI safety bill from Senators Amy Klobuchar, John Thune and Ted Cruz hit a new roadblock on September 23 when ranking member Maria Cantwell blocked it over a fundamental dispute about who performs mandatory safety evaluations of frontier AI models before deployment.
- Fact. The Klobuchar-Thune-Cruz draft lets companies run their own tests and present results to the Commerce Secretary for approval; Cantwell wants mandatory evaluation by federal agencies including national laboratories and national security bodies.
- Stake. With the EU’s AI Office already conducting its first compliance inspections, the Senate deadlock leaves the United States without a federal AI safety framework even as the world’s most capable models continue to scale, and as the labs themselves lobby for a patchwork of antitrust exemptions that further divides Congress.
The Senate Commerce Committee was expected to begin a markup session on September 23 for an AI safety bill co-sponsored by Klobuchar (D-MN), Thune (R-SD) and developed with input from Chair Ted Cruz (R-TX). Instead, the session was effectively blocked before it began when Cantwell’s office communicated her opposition to the bill’s central enforcement mechanism, according to people familiar with the matter who spoke to Nextgov. The markup had already been delayed once, after Cruz cancelled a planned session before the August recess to concentrate on child online safety legislation.
The Core Dispute: Who Tests, and How
The gap between the two versions is not marginal. The Klobuchar-Thune-Cruz bill would require frontier AI developers to conduct their own safety evaluations, submit findings to the Commerce Secretary, and receive approval before deploying a model. Critics describe this as a “primarily voluntary standard” — companies determine the scope of their tests and the bar they set for acceptable risk.
Cantwell insists that genuine safety evaluation of systems capable of “catastrophic risk” cannot be left to the developers themselves. Her alternative would require models to be assessed by federal agencies, including the Department of Energy’s national laboratories and relevant national security bodies, before deployment is permitted. One aide summarised her position bluntly: she wants “mandatory vetting,” not an industry self-certification process that mirrors how the financial sector’s internal risk models were treated before 2008.
Antitrust Waivers Add Another Fault Line
The safety-testing standoff sits alongside a separate but related dispute over antitrust exemptions. A bill authored by Senators Adam Schiff (D-CA) and Jim Banks (R-IN) would grant AI labs a limited exemption from antitrust law to share threat intelligence and coordinate responses to Chinese model distillation, cybersecurity incidents and other national security risks. Frontier labs have lobbied hard for the measure, arguing that safety coordination requires legal protection to be meaningful.
Senator Josh Hawley (R-MO) has emerged as its most vocal opponent. At a recent Senate hearing, Hawley said “there is no world” in which he would agree to give “the most powerful companies in the history of the world” an exemption from antitrust laws to collaborate, warning that such a carve-out would effectively let them collude under a safety pretext.
The two debates — who tests AI models, and whether labs may coordinate without antitrust risk — have become entangled, creating a legislative logjam that the House’s departure for campaign recess has done little to relieve. Meanwhile, the EU AI Office has already launched its first compliance inspections under the EU AI Act, targeting high-risk applications in HR, banking and healthcare. The contrast between Brussels’ enforcement posture and Washington’s deadlock has begun to surface in industry lobbying materials as an argument that US firms will face a patchwork of state regulations in the absence of federal action.
What Happens Next
Cruz has signalled he wants a bill ready before the year-end legislative calendar closes, but the window is narrow. The Commerce Committee has no confirmed markup date as of September 24. If the Klobuchar-Thune-Cruz bill cannot bridge the gap with Cantwell, Cruz may move a narrower version addressing only the antitrust exemptions or only the liability framework — neither of which satisfies the full scope of what advocates consider necessary before the next generation of frontier models reaches market.