← Back to feed

The White House signed an AI executive order on June 2 after weeks of cancellations and redrafts. The final version is materially smaller than what was proposed in May.

The core mechanism: AI companies are asked to submit powerful new models to a voluntary government review 30 days before public release. That window gives federal agencies time to assess threats the products may pose to financial systems, national security infrastructure, and other sensitive networks. The order is voluntary — no enforcement mechanism is specified in the signed text.

The 90-to-30-day collapse

The original draft, reviewed and signed off by senior White House officials and pre-cleared with OpenAI, Anthropic, and Google, called for a 90-day voluntary review window. AI companies lobbied against it, calling the period too onerous. On May 21, hours before a planned Oval Office signing ceremony with executives present, former AI czar David Sacks warned Trump that the order would slow innovation and disadvantage the US against China. Trump rejected the draft and postponed.

The redrawn order trims the window to 30 days and narrows the framing. Where the earlier version was a broad AI oversight measure, the signed order is explicitly cybersecurity-focused. It frames the directive as part of an “America First cybersecurity effort” aimed at maintaining AI dominance.

What it does and does not do

The order asks companies to cooperate — it does not compel them. There is no penalty structure for labs that release without completing the review, and no definition of which models qualify as sufficiently “powerful” to trigger the 30-day window.

For the major labs, the practical effect is likely minimal: all three that saw the earlier draft already have voluntary government engagement programs. Google, Microsoft, and xAI have also signed separate agreements under the CAISI pre-release testing framework. Anthropic’s Mythos Preview has had NSA and federal agency access since April.

What the order does establish is a stated federal preference for early government access to frontier models before commercial launch. That framing sets a precedent even in the absence of enforcement — future orders, legislation, or agency interpretations can reference it as the baseline.

Industry context

The sequence of events around this order illustrates the limits of federal AI policymaking under the current administration. A 90-day window became 30 days; binding obligations became voluntary asks; broad oversight became a cybersecurity brief. The labs got most of what they wanted while the White House retained the announcement.

The New York RAISE Act — state-level, with $1M penalties and a 72-hour incident reporting window — remains on track for January 2027 enforcement. The federal order does not preempt state-level AI regulation, a gap that continues to accumulate legal and compliance complexity for labs operating across US jurisdictions.

Key facts

• Signed: June 2, 2026 (privately, no public ceremony)
• Review window: 30 days (down from 90 days in canceled May draft)
• Compliance: voluntary
• Focus: cybersecurity threats from AI to financial, national security systems
• Previous draft: canceled May 21 on Sacks’ intervention
• No preemption of state AI laws