The Trump administration is exploring a framework that would hand the Pentagon responsibility for vetting artificial intelligence models before they reach federal, state, and local government agencies, according to sources familiar with White House deliberations.
The Office of the National Cyber Director convened two separate meetings last week, bringing together tech companies and industry trade groups to hash out security risks tied to advanced AI systems. Those discussions centered on a potential Pentagon-led testing regime designed to catch vulnerabilities before deployment across government.
The move marks a notable shift in posture. The administration revoked the Biden-era AI executive order on its first day, yet officials are now revisiting many of the safety frameworks that order contained. The Commerce Department's Center for AI Standards and Innovation would have handled such testing under Biden's approach; the Trump team is considering consolidating that authority with Pentagon oversight.
The timing is no accident. Anthropic's recent Mythos Preview release sparked fresh alarm about AI hacking capabilities and exposed gaps in the government's readiness to assess emerging model risks. Multiple agencies are now scrambling to understand what Mythos and OpenAI's GPT 5.5 can actually do, and what safeguards should exist before these tools reach government networks.
A White House official declined to confirm specifics, saying any policy moves "will come directly from the president" and dismissing talk of executive orders as speculation. But sources say the framework is substantially developed, with some work predating the Mythos uproar.
The administration faces competing pressures internally. Tech and economic policy voices worry that stringent testing requirements could slow deployment and complicate business operations. The national security establishment, by contrast, is focused on the possibility of an AI-enabled cyberattack on critical infrastructure or government systems.
The talks remain fluid. Proposals shift as different factions negotiate their positions, and sources say "real tensions" persist over how aggressively to regulate AI rollouts.
The White House is also weighing separate executive actions to allow federal agencies to circumvent an existing ban on using Anthropic, clearing a path for government use of Mythos if officials decide the security case is sound. That adds another layer of complexity to the emerging policy, since lifting restrictions would happen in parallel with new vetting requirements.
The core challenge facing policymakers is straightforward: move fast enough to capitalize on AI capabilities without creating openings for large-scale cyberattacks. Getting that balance wrong could expose government networks or critical systems to serious harm.
Author James Rodriguez: "The Pentagon has always been conservative about new tech, and handing it AI gatekeeping power is a smart hedge against the cybersecurity crowd's valid worst-case scenarios."
Comments