Sen. Mark Kelly endorses Trump administration push for Anthropic to pause access to top AI models

Sen. Mark Kelly (D-Ariz.) said Sunday that he agrees with the Trump administration’s decision to urge Anthropic to suspend access to its latest artificial intelligence models. Kelly’s remarks came as the administration sought to limit use of the company’s newest capabilities while the government evaluates how such systems are deployed and monitored.

Kelly said Anthropic’s models, including “Mythos and some of their other models,” are “incredibly capable,” and that restrictions on access are warranted because of the capabilities such systems can provide. He did not describe any criminal case or specific enforcement action, and his comments focused on the practical risks of deploying highly advanced AI without adequate guardrails and oversight.

The White House action referenced by Kelly fits into a broader pattern of executive-branch scrutiny of frontier AI development, particularly where officials contend that certain capabilities can be used for harmful purposes. The administration’s approach, as described by The Hill, centered on urging Anthropic to suspend access to its newest models rather than on a court order or legislation.

Kelly’s position also marks an unusual policy convergence between a senior Senate Democrat and President Trump on an issue that has previously divided parties. In public statements cited by The Hill, Kelly portrayed his agreement as grounded in the level of capability associated with the models and the need for responsible handling of technology he described as unusually powerful.

Under the model described by The Hill, Anthropic’s access decision would be voluntary in the sense that it is prompted by an executive-branch request. That means the immediate legal effect depends on what the company chooses to do, the scope of any pause, and whether the administration and the company negotiate terms or timelines for resuming access.

The situation is likely to raise questions for lawmakers and federal agencies about which authorities govern frontier AI systems and what standards apply to limiting public or customer access. For senators, the episode also highlights the ongoing tension between encouraging innovation and establishing timely constraints where officials believe risks are elevated, especially when systems are capable enough to do more than routine tasks.