Trump Administration Moves to Test AI Models Before Public Release
CNBC reported Tuesday that the Trump administration is deepening its AI oversight reach, signing evaluation agreements with Google DeepMind, Microsoft and Elon Musk’s xAI. The deals allow the federal government to assess those companies’ models before they reach the public.
Federal Testing Body Adds Three Major AI Players
The agreements were announced by the Center for AI Standards and Innovation, known as CAISI. The body sits within the U.S. Department of Commerce. Under the new deals, CAISI will run pre-deployment evaluations and targeted research on frontier AI systems to assess security risks and measure capabilities. Commerce Secretary Howard Lutnick has issued directives shaping the scope of that work.
Tuesday’s announcement extends a pattern established in 2024. CAISI struck earlier partnerships with OpenAI and Anthropic that year. Those existing agreements have since been renegotiated to align with Lutnick’s priorities and with the administration’s broader America’s AI Action Plan.
Also Read: What the America’s AI Action Plan Means for Big Tech
A Working Group Could Formalize the Process
Beyond the CAISI announcements, the White House is separately weighing a new AI working group, CNBC confirmed through a source close to the discussions. The group would bring together senior tech executives and government officials. Its mandate would include exploring oversight procedures and developing a framework for vetting models before release. One possible mechanism for establishing the group is an executive order, the source said. The source requested anonymity given the sensitivity of the talks.
The White House pushed back on specifics, telling CNBC that discussion of potential executive orders amounts to speculation. Any formal policy announcement, it said, would come directly from President Donald Trump.
Anthropic’s Powerful New Model Catalyzed Attention
The push toward formal oversight gained urgency last month. Anthropic unveiled a new model called Claude Mythos Preview, which demonstrated a sophisticated ability to identify software vulnerabilities and security flaws. The company limited its rollout to a select cohort of firms under a cybersecurity program called Project Glasswing.
Anthropic CEO Dario Amodei met with senior Trump administration officials at the White House shortly after the announcement. The meeting occurred despite the Defense Department having flagged Anthropic as a supply chain risk, a designation suggesting the company poses a potential national security concern. Both sides described the session as productive.
The widening circle of AI testing agreements signals that Washington is moving toward a more structured oversight posture. Whether a formal working group or executive order follows remains to be seen.
Read Next: Palantir Tops Estimates on 85% Revenue Growth in Fastest Expansion Since IPO
