WASHINGTON — A bipartisan Senate effort to regulate artificial intelligence remains stalled over how to hold AI companies responsible for dangerous systems and whether federal rules should override stronger state protections. The legislation, being developed by Sen. Amy Klobuchar, D-Minn., and Senate Majority Leader John Thune, R-S.D., would focus on what lawmakers describe as catastrophic AI risks. Draft language circulated in recent days has not been finalized, according to people familiar with the negotiations. Under the proposal, AI developers would have a legal duty to reduce harms caused by their models. The draft identifies risks including the creation of biological weapons and the use of AI to conduct offensive cyber operations. Companies that discover advanced capabilities in a model would have to notify the Commerce Department before releasing it. The disclosure would include the steps taken to address the risks and information that would allow the commerce secretary to verify whether those safeguards are sufficient. The department would not automatically approve every model. Instead, if the secretary concluded that a serious risk had not been adequately addressed, the department could seek an injunction in federal court. Without such an order, the company could decide whether to release the system. That approach has drawn opposition from Democrats who fear the proposal could weaken existing state safeguards. Sen. Maria Cantwell of Washington, the top Democrat on the Senate Commerce Committee, said a federal standard should not become “a backdoor for wiping out stronger state protections.” She also called for AI systems to be tested by scientists and other experts at national laboratories. The draft would preempt some state AI laws, a provision that has become one of the central points of disagreement. States and local governments have increasingly pursued their own rules for automated decision-making, consumer protection and other AI-related risks as Congress has struggled to act. Republicans, including Senate Commerce Chair Ted Cruz of Texas, have generally favored a less restrictive regulatory approach. Cruz has argued that excessive rules could slow American innovation and give China an advantage in the global AI industry. “It is possible to move bipartisan legislation,” Cruz said, “but it ain’t easy.” Klobuchar said she is continuing to pursue a bipartisan agreement and supports requiring developers to work with government experts to test and verify AI models. “We cannot allow the release of dangerous models, including those that evade the control of their developers,” she said. The negotiations accelerated after warnings from Jacob Coxon, a former Anthropic employee, that rapidly advancing AI could pose an existential threat within the decade. Rep. Sam Liccardo, D-Calif., separately urged House Speaker Mike Johnson to keep the House in session until Congress adopts AI safeguards. For communities and consumers, the unresolved debate will determine whether AI companies face enforceable obligations before releasing increasingly capable systems, or whether oversight remains divided among federal agencies, courts and state governments. The Senate has not set a clear timetable for completing the bill.