President Donald Trump cancelled a planned AI executive order shortly before its expected signing, saying he did not want oversight to weaken the United States’ competitive lead over China. Reports indicate that Elon Musk, Mark Zuckerberg and venture capitalist David Sacks spoke directly with him during the preceding night and morning.
The abandoned proposal was limited. It would have created a voluntary route for developers to engage federal agencies and submit advanced models for security review up to 90 days before release. It did not impose licensing or a mandatory delay.
Its cancellation reveals how even modest federal coordination has become entangled with industry influence and geopolitical competition.
A voluntary review became an innovation test
Trump described aspects of the order as a potential blocker without specifying which provisions caused concern. The argument that prevailed reportedly came from the accelerationist wing of the administration, which prioritises rapid capability and deployment.
The decision does not end every federal evaluation effort. The Centre for AI Standards and Innovation has agreements with Google DeepMind, Microsoft and xAI that allow government assessment before models reach the public.
However, the United States still lacks comprehensive AI legislation. Existing governance consists of agency guidance, executive actions and voluntary commitments that can change quickly with leadership and lobbying.
Earlier in May, the administration proposed a national legislative framework urging Congress to pre-empt state laws it considered unduly burdensome. Congress has not established the single federal standard the framework envisaged.
Direct access gives industry leaders policy power
Musk’s xAI competes with OpenAI and Anthropic, while Zuckerberg has positioned Meta around open model development. Both have commercial and strategic reasons to resist rules they believe could slow releases or advantage rivals.
Sacks had left his formal White House advisory role but reportedly retained enough influence to shape the decision. The episode suggests that a small group of executives can exercise an effective veto when they gain direct access to the president.
That influence raises an accountability problem. Frontier companies hold essential technical knowledge, so policymakers need their input. The same companies also benefit financially from the scope and timing of regulation.
OpenAI reportedly has White House support for a separate effort concerning state-level policy, even as the administration discourages fragmented state regulation. These parallel positions make it difficult for businesses and citizens to identify a coherent national approach.
Competition with China cuts in both directions
Trump framed cancellation as protection for the US lead. China, meanwhile, is accelerating formal AI legislation, requiring internal ethics review committees and placing governance language into state planning.
Faster US deployment may improve innovation and commercial scale. Regulatory uncertainty can impose its own cost by creating incompatible state rules, weak public confidence and unclear expectations for model security.
The comparison is not a simple choice between Chinese regulation and American freedom. Both countries are deciding how to manage increasingly autonomous and cybersecurity-relevant systems without suppressing useful development.
A voluntary federal review could have improved shared understanding of model risks without creating a licensing regime. Critics might still reasonably question confidentiality, scope and whether informal cooperation could become pressure to delay a release.
Those questions deserved a transparent policy debate. Cancelling the order after private conversations moves the decision away from public evidence and into relationships that outsiders cannot evaluate.
The United States can pursue speed and credible oversight at the same time, but doing so requires stable institutions, clear authority and safeguards against regulatory capture. A government evaluation programme must be technically competent enough to add value and independent enough to challenge the companies it reviews.
The cancelled ceremony leaves the underlying problem untouched. Frontier AI is becoming more capable, states remain active and federal legislation remains absent. Without a durable process, the next policy may again depend less on its substance than on which executive reaches the president before it is signed.
