White House Prepares Executive Order Requiring Government Early Access to Frontier AI Models
President Donald Trump is expected to sign an executive order this week prompting AI labs to grant government access to models 90 days prior to their release. The order also revises federal sharing of cyber threat information.
United States President Donald Trump is set to sign an executive order concerning artificial intelligence and cybersecurity, with the signing anticipated for this Thursday, May 21, according to Bloomberg. Axios published an exclusive report on May 20 outlining the content of the decree, which structures two distinct pillars: one aimed at early government access to frontier AI models and another dedicated to modernising federal threat intelligence sharing programs.
The decree will not mandate prior federal approval for the launch of AI models, a central point of contention within the White House in the weeks leading up to the signing. The framework adopted is voluntary: laboratories will be asked to notify the government and provide early access, but will not require formal authorisation for commercial release.
The Pillar of Access to Frontier Models
According to the structure described by Axios, developers of frontier AI models will be asked to provide pre-launch access to the government 90 days before any public availability. This access would also extend to operators of critical infrastructure, such as banks and hospitals, who would receive the models prior to broad commercial release for systemic risk assessment purposes.
The decree formalises and expands the security assessment programme of the Centre for Standards and Innovation in Artificial Intelligence (CAISI), a unit of the National Institute of Standards and Technology. CAISI had already established pre-launch assessment agreements with OpenAI and Anthropic in 2024. On May 5, 2026, Google DeepMind, Microsoft, and xAI signed similar agreements. The new executive order consolidates this framework into federal policy and extends the requirements to other laboratories developing models above certain computational capacity thresholds.
The Cybersecurity Pillar
The cybersecurity component of the decree reformulates federal threat information sharing programmes to include AI companies alongside sectors already integrated into these mechanisms, such as telecommunications and financial services. The measure also envisions strengthening the recruitment process for specialised personnel in federal agencies, focusing on skills for defence against AI-enhanced attacks.
According to Axios, the decree aims to bolster the security posture of critical infrastructure organisations, including hospitals, banks, and energy distributors, which will begin receiving technical support and threat intelligence directly from federal agencies, with AI companies integrated into this data-sharing network.
What Was Left Out and the Internal Debate
The final version omits the requirement for mandatory federal approval for AI models, a point advocated by factions within the White House concerned about national security risks. The opposing camp, strongly backed by industry stakeholders, argued that approval requirements would create unprecedented barriers to AI development in the United States. The adopted compromise, a voluntary regime with pre-launch access windows, represents a balance between the two positions.
Leaders from the technology sector have been invited to the signing ceremony at the White House. The presence of key AI laboratories at the ceremony will be interpreted by the market as a tacit endorsement of the voluntary regime adopted, signalling that the industry prefers direct negotiation with the federal government over the litigation of potential stricter regulations.
Implications for CIOs, CISOs, and Procurement Teams
For technology managers in companies integrating frontier AI models into critical operations, the decree establishes a new regulatory benchmark in the United States. The 90-day window for government assessment, if transitioned to a mandatory regime in future versions, would alter the planning schedules for AI deployment in sectors such as banking, insurance, and healthcare, where procurement windows are already defined by annual budget cycles.
For IT consultancies with federal contracts or clients in critical infrastructure sectors, the decree anticipates increasing demands for certification and due diligence of AI suppliers. Procurement teams that have not yet included clauses for access to government security assessment reports in contracts with suppliers of frontier models should review this position before the end of the third quarter of 2026, when the first compliance audits stemming from the decree are expected to be structured.