Trump Delays AI Executive Order on 90-Day Model Disclosure Rule

President Trump has delayed signing an executive order that would have established a voluntary framework for AI labs to share frontier models with the federal government 90 days before public release. The signing, originally scheduled for Thursday, has been pushed back as the White House works through final negotiations with AI companies and internal advisors.

The administration's stated reason: Trump told reporters he doesn't want to slow down AI development. The unofficial reason, per multiple reports: not enough major tech CEOs could get to Washington on short notice.

What the Order Would Do

The draft executive order has two main components, according to reporting from Axios, CNN, and The Next Web.

The first is a "covered frontier models" framework. AI labs above a certain capability threshold would be asked — voluntarily — to share unreleased models with the federal government and certain critical-infrastructure operators at least 90 days before public launch. The 90-day window would let government cyber and biosecurity teams probe the models for misuse risks before they ship.

The second is a cybersecurity component. The Treasury Department, alongside other agencies, would form a "clearinghouse" with AI companies to find and patch security vulnerabilities in frontier models before release.

The administration has indicated that OpenAI, Anthropic, Google DeepMind, Meta, and Microsoft are all expected to participate from the launch of the program.

Why this matters: A 90-day pre-release window would be the most significant new AI policy instrument in the U.S. since the Biden-era executive order, which the Trump administration revoked in early 2025. Even as a voluntary framework, it would reshape how labs plan model launches.

Why the Delay

Trump's public framing on the delay was characteristic. "I don't want to get in the way of that leading," he told reporters, referring to U.S. AI dominance.

The on-the-ground story, per Axios and CNN, is messier. Internal White House factions disagree on how prescriptive the order should be. MAGA-aligned national security advisors have pushed for tighter controls, citing concerns about Chinese and adversarial use of frontier models. Pro-industry advisors and several tech CEOs have pushed back, arguing that a 90-day pre-release review would slow U.S. competitiveness.

There's also a scheduling problem. The signing ceremony was meant to include major tech CEOs as a show of voluntary participation, and several couldn't make it to Washington on short notice.

The Industry Response

Frontier labs have been measured in public. OpenAI, Anthropic, Google, Meta, and Microsoft already share early model access with select government partners under existing voluntary arrangements. Formalizing that into an executive order standardizes what was previously bespoke.

Behind the scenes, the bigger sticking point is the definition of a "covered frontier model." Labs want the bar set high enough that not every fine-tune triggers a 90-day review. The government wants the bar set low enough to catch genuinely powerful new systems.

Industry sources flagged two practical concerns. First, a 90-day pre-release window is a long time in a market where competitors ship every few months. Second, the order does not include statutory protections, so a future administration could turn the voluntary framework into a mandatory one.

What's Already in Place

The current landscape is a patchwork. The U.S. AI Safety Institute, established under Biden, was rebranded and refocused under Trump. Labs voluntarily share certain pre-release evaluations with AISI and with the U.K. AI Safety Institute. State-level laws — most prominently California's SB 53 — impose their own disclosure and incident-reporting rules.

The new executive order is intended to consolidate these arrangements into a single federal framework. Whether that consolidation tightens or loosens the existing patchwork depends on the final draft language.

Industry Implications

For frontier labs, the delay buys time to negotiate the threshold language and the operational details of the 90-day review.

For state regulators, the executive order — if signed in something close to current form — could partially preempt state-level disclosure regimes. California, New York, and Colorado have been the most aggressive states on AI disclosure, and a federal voluntary standard could become a ceiling rather than a floor.

For smaller AI labs and open-weight developers, the most important question is the definition of "covered frontier models." If the threshold is set at the largest frontier systems only, the order has limited reach. If it's set broadly, it could affect open-weight releases and academic projects.

Expert Perspectives

Policy experts on X have been split. AI safety advocates argue a 90-day pre-release window is the minimum needed for serious testing of dangerous capabilities. Industry voices have pushed back, arguing that a 90-day clock effectively gives the government a veto window that could become a soft ban.

Several former White House officials have noted that voluntary frameworks tend to harden over time, especially after a high-profile incident.

What's Next

The White House has not given a new signing date. Reporting suggests the order could be signed within the next two weeks, once the language is finalized and a viable CEO showing can be coordinated.

Watch for three things: the final definition of "covered frontier models," whether the 90-day window is firm or adjustable, and which AI companies show up in person to sign the voluntary commitments alongside the president.

Bottom Line

The Trump AI executive order is a real policy instrument, not a press release. The 90-day pre-release disclosure framework would be the most significant new federal AI rule in two years. The delay isn't a sign the order is dead — it's a sign the negotiation is real, and the next two weeks will determine whether the U.S. has a federal voluntary AI disclosure standard or continues with the patchwork of state and agency-level rules.