5/19/2026, 1:00:00 PM · defense-government

Trump White House preparing AI executive order requiring 90-day pre-launch model disclosure to NSA

Draft order would create a voluntary framework for frontier AI developers to share models with federal agencies, including the National Security Agency, up to 90 days before public release.

The Trump administration is preparing an executive order that would establish a federal pre-release review process for advanced AI (Artificial Intelligence) systems, marking the most significant US regulatory move on frontier AI since the rescission of Biden-era policies. <cite index="6-1,6-2">Voluntary pre-deployment testing under the order would give government officials an opportunity to evaluate advanced AI models for cyber-related risks before they are broadly released, including whether the systems can assist with vulnerability discovery, as the administration grapples with the national security implications of advanced cyber-focused AI models.</cite>

<cite index="6-8">Axios first reported details about the order.</cite> <cite index="6-5,6-6,6-7">An NSA spokesperson referred questions to the White House, where an official said any policy announcement would come directly from the president and that discussion of potential executive orders is speculation.</cite>

Structure of the draft

According to officials and industry representatives briefed on the plans, the draft is organized around two pillars: a covered-frontier-model review track and a cybersecurity coordination track. <cite index="9-16,9-17">The Trump administration has drafted an executive order that would create a voluntary framework for AI companies to share their most advanced models with federal agencies before public release. Under the draft, companies developing frontier AI models would be expected to notify government agencies, including the Office of the National Cyber Director and the Cybersecurity and Infrastructure Security Agency (CISA), up to 90 days before making those models publicly available.</cite>

<cite index="9-5,9-6,9-7">The executive order is framed around cybersecurity rather than broad AI regulation. The key agencies named in the draft are the Office of the National Cyber Director, CISA, and the Office of Science and Technology Policy. Together, they would serve as the receiving end of any voluntary disclosures from AI developers.</cite>

Reporting indicates the National Security Agency (NSA) would also play a central evaluative role, with the Director of NSA making determinations on which systems qualify as "covered frontier models" in consultation with the National Cyber Director, CISA, and other officials.

Industry positioning

<cite index="4-17,4-18">The order is expected to include a voluntary agreement in which AI companies would share advanced models with the government for a period of time ahead of launch. The timeframe has been a point of discussion between the industry and the government, with one version of the executive order laying out an up to 90-day pre-launch review period while some of the AI companies involved prefer a shorter period, such as 14 days.</cite>

<cite index="9-4">Major AI laboratories, including OpenAI and Anthropic, have already been briefed on the proposed process.</cite> Reporting also indicates that Google DeepMind, Meta, and Microsoft are expected to be among the initial participants. <cite index="4-6">The Trump administration had taken a more hands-off approach to AI regulation until recently, when Anthropic unveiled its Mythos model, which it says can exploit cybersecurity vulnerabilities at an unprecedented pace.</cite>

Context

The move represents a notable shift for an administration that, on taking office, rescinded the Biden-era AI executive order and pursued a deregulatory posture. <cite index="6-11">President Donald Trump's AI Action Plan released last summer called for multiple agencies to establish an AI Information-Sharing Analysis Center to promote sharing of AI-related security threat information across critical infrastructure sectors.</cite>

Bipartisan signals in Congress suggest some appetite for intelligence-community involvement in model review. <cite index="6-12,6-13">Rep. Jim Himes, D-Conn., the top Democrat on the House Intelligence Committee, said it would be "insane" for U.S. intelligence agencies to not have early access to advanced artificial intelligence models that could be used for hacking and cyberdefense. He added that the Commerce Department should also play a role in the effort.</cite>

The White House has not published a final text, and the participating-company list, governance mechanism for the 90-day window, and timeline for the first model evaluated under the framework remain unconfirmed.

Cross-references

Sources

  1. [1]
    Trump to sign an AI oversight order this week as MAGA security pressure mounts on frontier-model labs
  2. [2]
    Trump's AI cybersecurity order: what the shelved draft actually says
  3. [3]
    Trump Abandons 'FDA for AI' Proposal | TechPolicy.Press
  4. [4]
    White House postpones executive order on AI | CNN Business
  5. [5]
    Here’s the Executive Order on AI That Gave Trump Cold Feet
  6. [6]
    Anticipated executive order could give NSA a role in voluntary AI model testing - Nextgov/FCW
  7. [7]
    Trump delays executive order on AI oversight hours before planned signing
  8. [8]
    US government’s AI executive order draft revealed by Politico, then pulled at the last minute
  9. [9]
    Trump administration's AI executive order shifts to voluntary model sharing with government
Trump White House preparing AI executive order requiring 90-day pre-launch model disclosure to NSA · AIDB