White House Puts OpenAI and Anthropic Models on a Short Leash Pending Cybersecurity Review
The Trump administration is vetting frontier AI releases before they reach the public — and both major labs are complying.
OpenAI confirmed Friday it is restricting access to its newest AI model at the explicit request of the Trump administration. Hours later, Anthropic followed with a similar announcement: its own latest model cleared a limited release only after receiving White House approval. Two of the most powerful AI labs in the world, gating their products at the direction of the executive branch.
The rationale is cybersecurity review. The administration is conducting what appears to be an unprecedented government vetting process for frontier AI systems before they reach commercial customers. Neither company disclosed the precise criteria being evaluated, the timeline for full release, or which agency is leading the assessment.
This is not a beat that usually overlaps with ransomware and criminal infrastructure — but it matters here. Frontier language models have become tools of interest to threat actors. Phishing lure generation, malware variant drafting, social-engineering scripts: these use cases are already documented in IR reports and RaaS operator forums. A government review process that slows broad access to the most capable models carries real implications for how quickly those capabilities spread beyond vetted commercial users and into the criminal ecosystem.
For now, access to both models flows through what amounts to an approved-customer list. The scope of that list — enterprise partners, government contractors, cleared researchers — has not been made public.
Compliance from both labs was swift. That speed is notable. OpenAI and Anthropic compete aggressively on capability benchmarks and talent acquisition. They don't typically move in lockstep. The fact that both accepted distribution restrictions within hours of each other suggests the administration's ask carried weight that neither company chose to test.
Whether this review process becomes a standing fixture for future model releases, or remains a one-time checkpoint, is the more consequential question.
