- The White House directed the Center for AI Standards and Innovation (CAISI) to stop publishing model assessments, citing national-security concerns as the rationale for the pause.
- National Cyber Director Sean Cairncross pushed to give security agencies more control over AI evaluation, a move that put him in conflict with White House AI adviser David Sacks, who worries excessive testing will slow innovation.
- The crackdown was triggered by the release of Anthropic’s Mythos and similarly powerful models capable of aiding cyberattacks or biological weapons development.
- OpenAI and Anthropic have lobbied to preserve CAISI’s independence, with OpenAI publicly calling for the unit to receive more resources and responsibility — not less.
What Happened?
The Trump administration has ordered CAISI — the Center for AI Standards and Innovation, housed within the Commerce Department — to cease publishing public evaluations of AI models. The directive came from National Cyber Director Sean Cairncross and other administration officials as they implement a new executive order Trump signed last week. CAISI, formerly known as the AI Safety Institute under Biden, is the primary government body that tests AI models before release and informs the public about their capabilities. While it continues to evaluate models internally, the suspension of its public-facing work has rattled AI companies and analysts who saw transparency as central to the unit’s mission.
Why It Matters?
CAISI sits at the intersection of two competing instincts inside the Trump administration: national security hawks who want to tightly control who can access the most powerful AI systems, and tech-aligned advisers who fear over-regulation will kneecap American AI leadership. The conflict has real consequences. Moving AI evaluation into classified or security-agency hands would effectively end the public transparency that companies like OpenAI and Anthropic have built their compliance programs around. It also risks creating a bottleneck: if security officials can slow or block model releases on national-security grounds, the deployment timelines of frontier AI systems — with enormous commercial implications — would become subject to an opaque, politically influenced process.
What’s Next?
The outcome hinges on how the new executive order’s implementation shakes out — specifically, whether CAISI is folded into the national-security apparatus, restructured, or eventually restored to its prior role. OpenAI’s public call to strengthen CAISI signals that leading developers will push back against a purely security-driven model. Cairncross and Sacks are expected to remain in tension, and the fate of CAISI’s leadership — following the abrupt resignation of a former Anthropic researcher who briefly ran the unit — remains unresolved. Congress may also weigh in, as AI oversight legislation is already circulating on Capitol Hill.
Source: The Wall Street Journal
