Trump Shelves AI Safety Order After Tech CEO Pressure, Exposing White House Split on China Strategy
President's last-minute cancellation of AI testing framework reveals deepening fracture between nationalist security hawks and Silicon Valley accelerationists as US-China AI competition intensifies.
President Trump abruptly cancelled a planned executive order on AI safety testing on May 21, hours before a scheduled Oval Office signing ceremony, after late-night calls from Meta CEO Mark Zuckerberg, xAI CEO Elon Musk, and AI adviser David Sacks convinced him the voluntary testing framework could function as an innovation ‘blocker’ against China.
The cancelled order would have created the first federal pre-release vetting system for frontier AI models, allowing agencies to identify cybersecurity vulnerabilities before public deployment. Trump’s reversal—framed publicly as protecting American competitiveness—signals a decisive victory for Silicon Valley’s deregulatory faction over nationalist security hawks within his coalition, according to Axios. “I didn’t like certain aspects of it. I postponed it,” Trump told reporters. “We’re leading China, we’re leading everybody, and I didn’t want to do anything to get in the way of that lead.”
“I really thought that could have been a blocker and I want to make sure that it’s not.”
— President Donald Trump
The Last-Minute Lobbying Campaign
Between Wednesday night and Thursday morning, Zuckerberg, Musk, and Sacks separately contacted Trump to express concerns about the proposed framework. Sacks—who stepped down as Trump’s AI and crypto czar in late March but retains significant influence—warned the voluntary vetting would function as a de facto licensing regime, slowing AI releases at a moment when Chinese competitors are narrowing capability gaps, according to Fortune.
The proposed executive order emerged from a sudden policy pivot triggered by Anthropic’s Claude Mythos model, released in April 2026. The model autonomously identified thousands of zero-day vulnerabilities in major software before public release, prompting the Trump Administration—which had spent January through April dismantling Biden-era AI safety initiatives—to reconsider its stance on pre-release testing, per Fortune.
The cancelled executive order would have established a two-part framework: immediate cybersecurity vetting of frontier models, followed by a broader voluntary testing system for AI models before public release. Unlike Biden’s November 2023 AI executive order—which Trump revoked in January 2026—the new framework was explicitly voluntary and focused narrowly on security vulnerabilities rather than broader safety concerns.
Coalition Fracture Exposes Strategic Vulnerability
The cancellation lays bare a deepening split within Trump’s coalition on AI governance. Nationalist populists including Steve Bannon have pushed for aggressive AI safety guardrails, while Silicon Valley libertarians led by Sacks, Musk, and Zuckerberg argue any testing requirement threatens U.S. competitiveness. Republican voter sentiment tilts sharply toward the nationalist position: 79% favor government testing of AI models for safety before public release, and 87% support government power to block releases posing National Security threats, according to May 22 polling by the Future of Life Institute cited by Fortune.
“It could be CEOs, or egos in general. Everyone hates each other in the political tech space,” an unnamed government official told Axios. The publication’s analysis warned that further delay “means more time for infighting and for the text to get bogged down in disagreements among different parts of the government and industry.”
- Trump’s reversal signals White House prioritises industry velocity over regulatory caution, potentially creating security blind spots
- Republican base support for AI safety testing (79%) directly contradicts administration’s accelerationist pivot
- Ad-hoc voluntary agreements with individual companies now replace coherent federal framework, limiting government visibility into frontier model capabilities
- China gains strategic advantage from US policy incoherence on AI governance
China Competition Drives Deregulatory Tilt
Trump’s rationale centres on maintaining U.S. lead over China in frontier AI capabilities. Yet the cancellation exposes a strategic paradox: Beijing’s centralised governance model enables rapid deployment of AI applications while Washington struggles to articulate a coherent leadership strategy that balances innovation with security, according to Modern Diplomacy.
The U.S. maintains advantages in semiconductor production and frontier model capabilities, but Chinese competitors including DeepSeek and Baidu have narrowed gaps significantly. DeepSeek recently shifted to domestic chip architectures following U.S. export restrictions, reducing dependence on NVIDIA hardware that had given Washington leverage, according to CGTN.
“The great fear in Washington is that China gets to AGI first,” Kyle Chan of the Brookings Institution told the Christian Science Monitor. Trump’s recent visit to China on May 12-14 underscored AI as the new centre of gravity in bilateral relations, with both capitals viewing artificial general intelligence as decisive for economic and military supremacy.
Ad-Hoc Agreements Replace Coherent Framework
In the executive order’s absence, the Trump administration has pursued voluntary testing agreements through the Commerce Department. Microsoft, Google DeepMind, and xAI signed individual agreements in early May for pre-release model vetting, but the patchwork approach limits government visibility into the frontier AI ecosystem compared to a unified federal framework, according to CNBC.
Notably, OpenAI and Anthropic—two labs most vocal about AI safety concerns—have not signed voluntary agreements, suggesting the ad-hoc strategy may exclude the very companies whose models prompted the administration’s initial safety pivot. Meta also remains outside the voluntary framework despite Zuckerberg’s direct lobbying against the executive order.
Casey K. Mace, senior official for APEC and economic policy, defended the administration’s approach in regional forums: “We’re very active in promoting U.S. AI options and solutions,” she told CNBC, though she did not address how voluntary agreements substitute for systematic pre-release testing.
What to Watch
Whether Trump attempts to resurrect the executive order in modified form, or whether the May 21 cancellation marks a permanent pivot toward industry self-regulation. Internal coalition tensions between nationalist security hawks and Silicon Valley accelerationists will likely resurface when the next frontier model incident exposes security vulnerabilities. China’s response to perceived U.S. regulatory paralysis—particularly whether Beijing accelerates deployment of AI applications in critical infrastructure—will test Trump’s bet that deregulation strengthens American competitiveness. Monitor which additional companies sign Commerce Department voluntary agreements, and whether OpenAI and Anthropic join the framework or remain outside it. Republican base polling showing 79% support for pre-release testing creates political vulnerability if a major AI security incident occurs before the 2028 election cycle.