President Donald Trump signed an executive order on June 2 requiring AI companies to voluntarily submit their most powerful models for government cybersecurity testing up to 30 days before public release, NPR reported. The White House framed the directive explicitly around a single geopolitical competitor: China. In doing so, the administration has made Beijing not just a backdrop to U.S. AI policy but its central organizing principle.
The order marks a meaningful, if contested, escalation in how Washington is attempting to manage the frontier AI development race. It also arrives at a moment when Chinese labs are pushing the boundaries of open-source development, domestic chip alternatives are advancing faster than most analysts expected, and the Trump-Xi summit produced only narrow diplomatic openings on AI safety. The executive order is Washington’s answer to the question of what comes next.
From 90 Days to 30: How Industry Lobbying Reshaped the Order
The signed order is a significantly trimmed version of an earlier draft that circulated in Washington before being shelved following sustained objections from prominent figures in the technology industry, including Elon Musk, Mark Zuckerberg, and AI policy advisor David Sacks. That original draft called for a 90-day pre-release review window, a timeline that AI developers argued would hand China a structural advantage by slowing American lab release cycles while Chinese competitors faced no equivalent constraint.
The criticism was pointed enough to kill the original draft entirely. What emerged in its place is a 30-day voluntary framework — a compromise that critics on one side call toothless and proponents on the other argue is still dangerously bureaucratic. The word “voluntary” carries enormous weight here. The administration has no formal enforcement mechanism to compel compliance, meaning the order functions more as a policy signal than a binding regulatory regime.
That signal, however, is not insignificant. By explicitly invoking China as the rationale for the order, the White House has embedded geopolitical competition into the formal architecture of American AI governance in a way that will shape future, potentially binding, legislation.
Why Beijing Is Watching Closely
From Beijing’s vantage point, the executive order is a double-edged development. On one hand, a weakened, voluntary-only review requirement represents a partial victory for the argument that heavy-handed U.S. regulation could slow American frontier labs — an argument Chinese officials and state media have made repeatedly. On the other hand, the explicit naming of China as the threat driving U.S. AI policy reinforces a bilateral competitive framing that Beijing has generally preferred to avoid in multilateral settings.
The 9th Digital China Summit in Fuzhou and the 2026 World Intelligence Expo in Tianjin both showcased the breadth of China’s AI ambitions — from foundation models to embodied robotics to 6G infrastructure — at a moment when Washington is still debating whether a 30-day review window is too long or too short. The contrast in cadence is deliberate and not lost on either government.
Chinese AI labs, meanwhile, have structured their development around a different kind of constraint: U.S. export controls on advanced semiconductors. That pressure has accelerated domestic chip development and pushed Chinese companies toward open-source model strategies that, ironically, face no equivalent review requirement anywhere in the world. Alibaba’s Qwen family recently surpassed one billion downloads, capturing over 50 percent of the global open-source AI market, a figure that American policymakers cite when arguing that open-source proliferation itself poses national security risks.
The Cybersecurity Testing Question
The specific focus on cybersecurity testing, rather than broader safety evaluations or capability thresholds — is telling. It suggests the administration is less concerned with the existential risk framing that dominated AI governance discourse in 2023 and 2024, and more focused on near-term vulnerabilities: models that could be exploited for cyberattacks, disinformation, or the acceleration of weapons development.
That framing aligns with the AI guardrails discussion at the Trump-Xi summit, where both sides acknowledged overlapping fears about AI-enabled cyber operations even as they disagreed on virtually every structural question about how to address them. The executive order can be read as the U.S. attempting to demonstrate domestic credibility on those shared concerns without conceding any ground on the competitive dimensions of the relationship.
Whether cybersecurity testing in a 30-day window can meaningfully catch sophisticated vulnerabilities in frontier models is a separate technical debate that the order does not resolve. Researchers have noted that red-teaming exercises of sufficient depth typically require more time and access than a voluntary pre-release review would provide.
A Policy Signal in Search of a Framework
The deeper problem with the executive order is structural. Voluntary compliance regimes in competitive industries tend to erode under commercial pressure, particularly when the companies being asked to comply are racing against foreign competitors who face no equivalent expectation. The original 90-day proposal, whatever its flaws, at least attempted to establish a binding norm. The 30-day voluntary version is closer to a handshake agreement between the government and an industry that successfully lobbied to weaken the original terms.
This is not unique to AI. The pattern of ambitious regulatory proposals being scaled back after industry intervention is familiar across sectors. What makes the AI case distinctive is the explicit national security framing. When the White House argues that pre-release review is necessary to stay ahead of China, it is simultaneously arguing that any weakening of that review — including the weakening it just accepted, creates a national security gap. That logical tension will not disappear.
For Beijing, the spectacle of Washington’s internal debate over AI governance is informative in itself. China’s regulatory approach to AI has been criticized as opaque and politically motivated, but it has not faced the same kind of public industry pushback that forced the Trump administration to halve its review window and remove mandatory compliance requirements. Whether that reflects a more coherent policy environment or simply a less pluralistic one is a question that cuts to the heart of how the two systems are competing.
What is clear is that the executive order, in its final form, represents neither the regulatory floor its original architects envisioned nor the non-intervention preferred by the Musk-Zuckerberg-Sacks coalition. It is a compromise document in a race that does not reward compromise, and Beijing will be paying close attention to what comes next.