Sebastian Mallaby, a senior fellow at the Council on Foreign Relations, said U.S. policy must control dangerous frontier artificial-intelligence models [1].
This push for regulation comes as the U.S. government begins treating advanced AI as a strategic security concern. Because these models can be misused, Mallaby said they should be managed similarly to other dual-use technologies that have both civilian and military applications [3].
Speaking on CNBC’s program “The Exchange,” Mallaby said the security risks associated with the most capable AI systems are significant [1]. He said the potential for misuse necessitates a controlled approach to deployment and access [3].
Recent government actions reflect this shift toward restriction. On June 13, 2026, the U.S. government ordered Anthropic to halt foreign access to certain advanced AI models [4]. Some reports attribute this specific directive to an order from the Trump administration [5].
Other reports indicate that U.S. Commerce officials have started applying restrictions similar to export controls on the most advanced models [5]. These measures aim to prevent high-capability AI from falling into the hands of adversaries.
The pace of development continues despite these regulatory efforts. For example, Anthropic debuted its Mythos model in April 2026 [3]. The emergence of such powerful tools underscores the tension between rapid innovation and national security requirements.
“U.S. policy must control dangerous frontier artificial-intelligence models”
The shift toward treating frontier AI as a dual-use technology signals a transition from viewing AI primarily as a commercial product to viewing it as a national security asset. By implementing export-style controls and restricting foreign access, the U.S. is attempting to maintain a technological lead while mitigating the risk that advanced models could be used to develop cyberweapons or other strategic threats.
