Trump’s New AI Oversight Order: What Changed

On June 2, the Trump administration signed a revised AI oversight order, cutting the voluntary pre-launch review window from 90 days to 30. The original, more demanding version was blocked in late May following coordinated industry opposition led by former White House AI czar David Sacks.

Key Takeaways

• The final order sets a 30-day voluntary submission window for advanced AI models, down from 90 days in the initial draft.
• The text explicitly states that nothing in the order authorizes mandatory licensing, preclearance, or permitting requirements.
• The DOJ is directed to prioritize AI-assisted crimes, including hacking and unauthorized system access.

From 90 Days to 30: How the Industry Won This Round

The executive order that had been circulating in Washington for weeks was not going to survive intact. By late May, Trump had already delayed a first signing after repeated warnings from the tech industry. The original draft required AI companies developing advanced models to submit them to the federal government for evaluation up to 90 days before any public release.

The industry’s target was roughly two weeks. The final number landed at 30 days. In practical terms, that sits far closer to what the industry wanted than what the original draft proposed.

The lobbying effort organized around one central figure: David Sacks, former White House AI czar turned venture capitalist. His position served as a rallying point for tech stakeholders opposed to the original text. Trump ultimately signed the revised version privately, without the Silicon Valley CEO photo opportunity that had originally been planned.

The argument that shaped the outcome was geopolitical as much as technical. Trump stated he did not want to impede US AI companies in their competition against Chinese players. The China competitiveness framing carried more weight than any technical argument about review timelines. Trump’s ties to the industry run deep, reportedly holding an equity stake in OpenAI.

The result: a voluntary 30-day submission window that, operationally, changes very little for the major AI labs. These companies already run internal evaluation cycles of comparable length before any significant launch. The compliance cost of this AI oversight order is effectively zero for any serious lab.

What the Order Actually Says

The final text establishes a voluntary submission process for advanced AI models in the 30 days prior to commercial release. The companies covered are those developing “advanced AI models,” a definition left deliberately broad. Submissions are intended for government evaluation. Not as a prerequisite for launch.

The most discussed clause is the one that explicitly removes any coercive power: “Nothing in this section shall be construed to authorize a mandatory governmental licensing, preclearance, or permitting requirement.” This language was inserted directly into the body of the order, addressing industry concerns about an FDA-style regime for AI development.

The Department of Justice receives a separate mandate: prioritize enforcement of AI-assisted crimes, particularly hacking and unauthorized system access. This targets offensive use of AI by malicious actors without touching the development processes of legitimate companies. The division of labor is clear: innovation remains unconstrained, prosecution of abuses falls to law enforcement.

As illustrated by the deployment of Claude Mythos across critical infrastructure in 15 countries, questions of AI governance in sensitive environments are now firmly on the global regulatory agenda. This executive order does not answer those questions. It deliberately sidesteps them.

The order continues the logic of Trump’s December 2025 executive order, which aimed to create a unified national “rulebook” to preempt state-level AI regulation. The thread is consistent: less federal oversight of AI development, regardless of the surface-level framing.

Also on Horizon:

• Alphabet Raises $80 Billion to Finance Its AI Infrastructure
• Claude Mythos Secures Critical Infrastructure in 15 Nations
• Codex Goes Enterprise: OpenAI Targets White-Collar Workers

What This Changes for AI Companies, Short and Medium Term

The watered-down executive order sets a clear political precedent: tech lobbying can shape US AI regulation, even under an administration that had signaled interest in stronger oversight. For companies preparing for public markets, including Anthropic, which recently filed confidentially with the SEC, this regulatory clarity arrives at a useful moment.

In the short term, major AI companies keep their release cadence intact. A voluntary 30-day window creates no real operational friction for players that already manage internal evaluation cycles of similar length before any major launch. Immediate impact on shipping schedules: none.

In the medium term, the US position may produce the opposite of its intended effect in Europe. The EU’s AI Act imposes significantly stricter obligations on high-risk models. European regulators could read this American retreat as validation of their own tougher approach. The transatlantic regulatory gap widens, creating deployment asymmetries for companies operating on both sides.

The DOJ mandate signals a broader political choice: the US government prefers to regulate harmful uses after the fact rather than constrain development upstream. This leaves the market to define deployment norms, with criminal enforcement as the only backstop against the most egregious cases.

The broader question is international spillover. The AI oversight order sets a US precedent that other governments are watching. Several countries within the G7 look to Washington as a regulatory reference point. An American retreat could give other jurisdictions the opening to impose their own standards first turning AI regulation into the next terrain of geopolitical competition.

Follow the story on Horizon.