Trump Signs Executive Order Giving Federal Agencies 30-Day Pre-Release Access to Frontier AI Models

President Donald Trump signed an executive order on Tuesday directing the federal government to seek up to 30 days of pre-release access to the most powerful frontier AI models — a significant policy reversal driven by growing alarm over AI's ability to discover and exploit software vulnerabilities at scale.

The order, titled "Promoting Advanced Artificial Intelligence Innovation and Security," tasks the departments of Homeland Security and Treasury, the White House Office of the National Cyber Director, and NIST with 60 days to establish criteria defining what constitutes a model powerful enough to require government review. Once those criteria are published, AI developers will be expected to identify qualifying models and notify the government, which can then request pre-release access lasting up to 30 days.

From Deregulation to Early Access

The order marks a notable policy reversal. On his first day in office in January 2025, Trump rescinded Biden's AI safety executive order, calling federal oversight requirements overly burdensome. The catalyst for the shift: Anthropic's Claude Mythos, a frontier model with advanced vulnerability-hunting capabilities that alarmed national security officials who fear such tools could be weaponized before defenses are in place.

A previous draft had called for a 90-day pre-release review window. Tech executives pushed back hard, arguing that timeline would stifle deployment. Trump pulled that version on May 21 following intense industry lobbying. Tuesday's signed order narrows the window to 30 days — a compromise that still drew criticism from Democratic lawmakers who noted the administration had previously rejected the very provisions now enacted.

What the Order Does

Beyond the pre-release access framework, the EO establishes several concrete programs:

• AI Cybersecurity Clearinghouse: A voluntary coordination mechanism between the government, AI developers, and critical infrastructure operators to identify and remediate software vulnerabilities at scale.
• Broadened tool access: DHS is directed to issue binding operational directives facilitating access to AI-enabled cybersecurity tools for federal agencies, state and local governments, and critical infrastructure operators including rural hospitals, community banks, and local utilities.
• Classified benchmarking: A classified process for evaluating frontier models' offensive cyber capabilities, providing a consistent yardstick for what qualifies as a covered model.
• Cybersecurity workforce expansion: OMB and OPM are directed to identify funding and expand federal cybersecurity hiring pathways.

The order explicitly states that nothing in it authorizes mandatory licensing, pre-clearance, or permitting requirements for AI development or distribution — language inserted directly in response to industry concerns about earlier drafts.

Who Gets Early Access

The voluntary framework targets "trusted partners" — select critical infrastructure operators who would receive early model access alongside government agencies to test defenses before public release. The arrangement is designed to help hospitals, utilities, and financial institutions identify AI-driven attack vectors before adversaries do.

Anthropic separately expanded access to its Claude Mythos Preview model to 150 additional organizations including critical infrastructure operators, signaling that the industry and government are moving in parallel on the vulnerability-testing front.

As reported by Cybersecurity Dive, the order represents the most direct federal intervention in AI development since the Biden administration — one the Trump team explicitly criticized just 18 months ago.