US Senators Thune (R), Cruz (R) and Klobuchar (D) were close to proposing a bill to deal with the risks of advanced AI, though disagreements with Anthropic put the agreement on hold, sources told Punchbowl

• Anthropic pushed to require public disclosures about risks from its systems; Anthropic also objected to giving the Commerce Department an ongoing injunction power.

Bipartisan AI bills in Congress have repeatedly stalled at the drafting stage, and the pattern here fits prior episodes: near-agreement between senior senators, then a hold-up over the enforcement architecture rather than the principle. The fault line reported, mandatory risk disclosure versus ongoing injunction powers for the Commerce Department, is the same divide that has separated the industry-preferred transparency regime from a regulator with standing intervention authority, and which side prevails historically determines whether compliance is a reporting cost or an operating constraint. A single frontier lab negotiating directly against the draft is unusual in degree but not in kind; large incumbents have previously shaped tech legislation by withholding support at the staff level, and their objections tend to mark the provisions that survive into later vehicles. Worth watching is whether the disagreement delays introduction into the next session or is resolved into a narrower disclosure-only bill, since the former path has typically meant the effort dissipates while the latter produces a template that states and agencies then copy. For the AI complex, the transmission runs through compliance burden and deployment timelines for frontier developers and their hyperscaler partners, not through near-term revenue. As a sourcing story about a bill that does not yet exist, the signal is directional.