White House Shelves Mandatory AI Testing Order After Industry Pressure
Trump postpones pre-deployment review framework following last-minute calls from Musk and Zuckerberg, leaving voluntary agreements as sole governance mechanism.
The Trump administration postponed a planned executive order requiring pre-deployment testing of frontier AI models on 22 May 2026, hours before signing, following calls from Elon Musk, Mark Zuckerberg, and AI czar David Sacks warning the mandate could handicap US competitiveness against China.
The shelved order would have formalised a 90-day voluntary review process for advanced AI systems before public release, framing safety oversight as a National Security imperative comparable to FDA drug approval. Kevin Hassett, Director of the National Economic Council, defended the framework in statements to Federal News Network: “We’re studying possibly an executive order to give a clear road map to everybody about how this is going to go and how future AI that also potentially create vulnerabilities should go through a process so that they’re released in the wild after they’ve been proven safe, just like an FDA drug.”
The decision exposes deep divisions within the administration over whether stricter oversight strengthens American AI leadership by establishing global governance standards, or undermines it by slowing deployment relative to less-regulated Chinese competitors. Per Axios, the draft order contained two sections: one mandating cybersecurity standards for federal AI procurement, the other establishing the voluntary pre-deployment review framework that ultimately triggered industry resistance.
voluntary framework now carries full governance burden
The postponement leaves voluntary testing agreements signed with the Commerce Advanced Intelligence Safety Institute as the sole federal governance mechanism. CAISI expanded these commitments on 5 May 2026 to include Google DeepMind, Microsoft, and xAI, building on prior 2024 agreements with OpenAI and Anthropic. Chris Fall, CAISI Director, stated the expanded partnerships “help us scale our work in the public interest at a critical moment,” according to CNBC.
The institute has completed over 40 pre-deployment evaluations of frontier models, including unreleased systems with state-of-the-art capabilities. But industry scepticism runs high. Dean Ball, primary author of the White House AI Action Plan, told The Hill: “If you needed yet more evidence that the burden of frontier AI governance is going to rest principally on the private sector, you got it yesterday.”
A former Trump White House official questioned the durability of voluntary commitments in the same report: “Despite it being voluntary, what is to say it stays that way?” The concern reflects broader industry wariness that today’s optional framework could morph into tomorrow’s regulatory mandate once testing infrastructure matures.
“They created the race, they’re running the race, and now they’re saying the race is forcing them to cut corners.”
— Gary Marcus, AI researcher, The Editorial
compliance costs already reshaping competitive dynamics
Cross-industry AI compliance spending averaged $5.2 million per firm in April 2026, with high-risk systems requiring additional certification that doubles costs, per SQ Magazine. Federal agencies introduced 59 AI-related regulations in 2024 alone — more than double the prior year — while 78% of organisations reported deploying AI systems, according to Credo AI data published in December 2025.
The compliance burden falls unevenly. Safety-focused labs like Anthropic already restrict access to advanced models based on internal risk assessments — the company’s Mythos system, announced in April 2026, identifies cybersecurity vulnerabilities at unprecedented speed but remains limited to approved organisations. Rivals deploying faster with lighter safety protocols gain market advantage, intensifying pressure to accelerate even within risk-averse teams.
Sam Altman of OpenAI acknowledged the competitive tension in remarks to The Editorial, stating “the race dynamic is real” and calling for international coordination to prevent a race to the bottom on safety standards. Critics see this framing as self-serving — Gary Marcus, an AI researcher, countered that frontier labs “created the race, they’re running the race, and now they’re saying the race is forcing them to cut corners.”
geopolitical stakes frame governance as strategic competition
The shelved executive order reflects broader uncertainty over whether pre-deployment testing strengthens or weakens US positioning against China. Analysis from Geopolitical Monitor frames the governance race as a new domain of strategic competition, with the US pursuing soft-power standards-setting while China positions state institutions for alternative frameworks.
Establishing robust testing protocols could cement American leadership by exporting governance norms globally — the template that EU regulation and allied governments adopt. Conversely, mandatory reviews could slow US deployment cycles while Chinese labs operate under lighter oversight, accelerating their catch-up trajectory.
Biden’s October 2024 National Security Memorandum framed frontier AI testing as a national security imperative using nuclear, space, and pandemic analogies, directing NIST’s AI Safety Institute to conduct voluntary pre-deployment evaluations. Trump initially continued this approach through expanded agreements but reversed course after industry pressure. The administration has simultaneously proposed a 30% budget cut to NIST, the agency responsible for developing AI safety standards, according to the Bulletin of the Atomic Scientists.
The timeline pressure is acute. Anthropic’s restricted Mythos deployment in April demonstrated AI-enabled cyber capabilities that alarmed government agencies. Simultaneous regulatory momentum from the EU AI Act, California SB 53, and Colorado SB 205 creates fragmented compliance burdens that could disadvantage smaller US labs unable to navigate multiple jurisdictions.
looking ahead
The durability of voluntary CAISI agreements will depend on whether frontier labs face sufficient deployment pressure to abandon commitments when competitive disadvantage materialises. State-level regulatory acceleration is likely to continue regardless of federal action — California and Colorado have advanced legislation despite Washington’s retreat, creating compliance fragmentation that larger labs can navigate more easily than startups.
International coordination efforts will determine whether US testing frameworks become global standards or regional outliers. If allied governments adopt CAISI-style protocols while China pursues lighter-touch oversight, the strategic calculus shifts toward governance as competitive handicap rather than leadership tool.
The administration’s next move remains uncertain. No new signing date has been announced for the shelved order, and industry sources suggest the voluntary framework may remain the ceiling rather than the floor for federal oversight. Whether self-regulation can survive commercial pressure when safety-focused labs already face deployment speed demands will test the viability of governance-by-agreement in a technology domain where national security and market dominance converge.