Story
July 1, 2026
U.S. Government to Review AI Models for National Security Risks
The U.S. government has struck agreements with Google DeepMind, Microsoft, and xAI to review their advanced AI models for national security risks before public release. The initiative, led by the Center for AI Standards and Innovation (CAISI), will assess potential threats related to cybersecurity, biosecurity, and chemical weapons.
The Trump administration’s latest move to pre‑screen cutting‑edge AI models for national security risks has united tech giants and regulators in a rare show of cooperation, even as it deepens long‑running worries about government overreach and opaque executive power.
Expanding federal reach into “frontier” AI
Under new agreements announced by the Center for AI Standards and Innovation (CAISI), a Commerce Department unit, Google DeepMind, Microsoft and Elon Musk’s xAI will allow the U.S. government to evaluate advanced AI models before public release.1 CAISI says it will conduct “pre-deployment evaluations and targeted research to better assess frontier AI capabilities and advance the state of AI security,” expanding earlier deals first struck with OpenAI and Anthropic in 2024.1
A parallel initiative, reported by CNBC, would go further: the White House is weighing a new AI working group that could formalize pre‑release vetting procedures and may even be created by executive order, according to unnamed sources.1
National security case: cybersecurity, biosecurity, and chemical risks
CAISI frames the deals as a response to concrete security threats. The agency describes the review process as “key to understanding the capabilities of new and powerful AI models as well as to protecting US national security,” emphasizing that independent, rigorous measurement science is essential to grasp “frontier AI and its national security implications.”2 The focus is explicitly on models that could bolster hacking, bioengineering, or chemical weapons development.2
Officials point to Anthropic’s Mythos model—so adept at finding software vulnerabilities that its rollout was tightly restricted and paired with a cooperative cybersecurity effort, Project Glasswing—as a warning sign of what unregulated releases might enable.12
Civil-liberties and power-concentration concerns
Yet the same mechanisms that reassure security hawks alarm civil‑liberties and competition advocates. Pre‑release testing through a Commerce‑run center, reinforced by a White House working group populated by major tech executives, risks entrenching incumbent firms as gatekeepers over which “frontier” models can safely launch.1 The reliance on confidential discussions and possible executive orders underscores how much of this emerging oversight regime may be shaped outside normal legislative debate, leaving transparency and democratic accountability trailing well behind the technology it seeks to control.