AI Safety Regulation Halted by Trump After Silicon Valley Lobbying

On the morning of Thursday, May 21, 2026, the wheels of federal policymaking ground to a sudden, screeching halt. High-ranking tech executives were already boarding flights to Washington, D.C., and invitations had been dispatched for a high-profile White House signing ceremony. The event was intended to debut a landmark executive order designed to establish a federal safety vetting framework for advanced artificial intelligence models. Yet, just hours before he was scheduled to put pen to paper, President Donald Trump abruptly withdrew the draft, leaving both national security hawks and tech policy advocates stunned. The sudden collapse of this executive order represents a watershed moment in the global battle over AI safety regulation, signaling that in the high-stakes clash between domestic risk mitigation and geopolitical dominance, the raw pursuit of innovation has emerged victorious.

Speaking to reporters in the Oval Office, Trump explained his last-minute change of heart in characteristic terms, framing the decision not as a retreat, but as a strategic defense of American exceptionalism. “I didn’t like certain aspects of it, I postponed it,” Trump stated. “We’re leading China, we’re leading everybody, and I don’t want to do anything that’s gonna get in the way of that lead.” Behind the scenes, however, this dramatic policy reversal was not a solo decision. It was the direct result of a coordinated, eleventh-hour lobbying blitz spearheaded by some of Silicon Valley’s most powerful figures, who leveraged the specter of a rising China to dismantle what they viewed as a regulatory chokehold on the next digital frontier.

Inside the Leaked Draft: What the Proposed AI Safety Regulation Actually Sought

To understand the magnitude of the industry’s lobbying victory, one must look at what the proposed executive order actually contained. A leaked seven-page draft, obtained and published by POLITICO, revealed a framework that attempted to walk a tightrope between national security anxiety and market freedom. Far from the draconian restrictions feared by tech libertarians, the draft order actually proposed a voluntary system rather than a mandatory licensing regime. It specifically outlined several primary pillars of federal interest:

• Cybersecurity Defense: The order sought to secure critical national infrastructure—including federal databases, electrical grids, hospitals, and financial institutions—against sophisticated, AI-driven cyber threats.
• Frontier Model Vetting: It established a voluntary “clearinghouse” under the federal government, inviting developers of “covered frontier models” to share their state-of-the-art technologies with federal agencies, such as the National Security Agency (NSA), for safety testing.
• The Pre-Release Window: AI labs were encouraged to submit their advanced models for evaluation up to 90 days prior to public deployment, giving intelligence agencies a chance to locate underlying security flaws and exploitation vectors.
• CFAA Enforcement: The draft directed the Department of Justice to aggressively enforce the Computer Fraud and Abuse Act (CFAA) against bad actors utilizing AI to compromise digital networks.

Crucially, the text of the draft explicitly prohibited the creation of a “mandatory governmental licensing, preclearance, or permitting requirement.” It was designed to coax, not coerce, companies into cooperating with Washington. Yet, even this soft-power approach was deemed an unacceptable bottleneck by the titans of Silicon Valley, who viewed any formal pre-release delay as a dangerous precedent that could eventually ossify into a bureaucratic “FDA for AI.”

The Mythos Shockwave: Why AI Safety Regulation Reached a Crisis Point

The impetus for the executive order was not theoretical; it was sparked by a profound technical “reckoning” that shook the defense and financial sectors. In April 2026, the artificial intelligence safety lab Anthropic announced the development of its latest model, Claude Mythos. However, in an unprecedented move, Anthropic chose to withhold the model from public release, sounding an alarm that reverberated through the halls of the Pentagon and Wall Street.

The technical capabilities of Claude Mythos represented a quantum leap in autonomous software execution. The model demonstrated an unprecedented, highly localized ability to discover unknown, “zero-day” security vulnerabilities in complex source code, and subsequently write autonomous scripts to exploit them. While developers envisioned the tool as a defensive shield for patching code, national security officials immediately recognized its potential as an offensive weapon of mass disruption if it fell into foreign hands. Anthropic’s self-imposed restriction triggered a quiet geopolitical crisis, as allies and adversaries alike realized that frontier AI was rapidly transitioning from a cognitive assistant into an autonomous cyber-agent.

In response to the “Mythos moment,” the Trump administration initially shifted into high gear. Treasury Secretary Scott Bessent and outgoing Federal Reserve Chair Jerome Powell convened emergency meetings with Wall Street CEOs to warn them of the systemic risks Mythos-class models posed to global financial transactions. Concurrently, the White House began exploring structural changes to its previously laissez-faire stance, considering plans to have the National Security Agency lead classified, pre-release evaluations of new models. The momentum toward federal oversight seemed unstoppable—until Silicon Valley’s ultimate power players intervened.

The Billionaire Intervention: How Tech Giants Dismantled the Deal

As details of the upcoming executive order leaked in mid-May, the backlash from the tech industry’s deregulatory wing was swift and severe. Between the evening of Wednesday, May 20, and the morning of Thursday, May 21, a sequence of high-stakes phone calls directly to President Trump systematically dismantled months of bureaucratic negotiation. The chief architects of this lobbying effort included SpaceX and xAI founder Elon Musk, Meta CEO Mark Zuckerberg, and Trump’s own White House “AI and Crypto Czar,” David Sacks.

David Sacks, who had previously championed a centralized national standard to preempt a patchwork of conflicting state laws, argued passionately that the proposed pre-release testing regime was a strategic blunder. Sacks warned Trump that even a voluntary 90-day review period would create immediate operational bottlenecks, slowing down the launch of American products at a time when speed is the ultimate metric of success. Furthermore, Sacks raised a potent political concern: while the current administration might keep the framework voluntary, a future, more regulatory-minded administration could easily weaponize the infrastructure of the executive order to implement mandatory pre-clearance rules.

Musk and Zuckerberg reinforced this narrative, appealing directly to Trump’s economic instincts. They argued that imposing a 90-day waiting period on American developers while foreign competitors operated with zero restrictions would cripple the U.S. economy. This message resonated deeply with Trump, who has consistently viewed the domestic AI sector as a primary engine of economic growth and a cornerstone of national security. By early Thursday afternoon, the signing ceremony was aborted, and the draft order was shelved.

The Geopolitical Game Theory: Why Competitiveness Trumps AI Safety Regulation

The abrupt postponement of the executive order highlights a fundamental truth of modern technology policy: in the era of great-power competition, the geopolitical race for AI supremacy against China acts as an absolute solvent against domestic regulation. Every effort to introduce safety guardrails is inevitably met with the argument that doing so will hand the crown to Beijing. Under this framework, AI safety regulation is viewed not as a shield to protect domestic infrastructure, but as a self-imposed handicap in a winner-take-all technological cold war.

This dynamic has created a complex paradox for federal policymakers. While intelligence agencies remain deeply concerned about the immediate, tangible threats of autonomous hacking agents like Claude Mythos, the executive branch has decided that the risk of falling behind China is far more dangerous than the risk of deploying untested AI. By prioritizing speed over safety, the administration has effectively embraced a policy of forward escape, hoping that American technological dominance will naturally yield defensive solutions to the very vulnerabilities these models create.

The Post-Order Landscape: A Greenlight for Unchecked Momentum

The cancellation of the May 21 executive order is a historic triumph for Silicon Valley’s deregulatory advocates, cementing the influence of a small cohort of tech billionaires over the highest levels of U.S. statecraft. For the foreseeable future, the development of frontier artificial intelligence will remain a self-policed endeavor, characterized by rapid, hyper-competitive releases and minimal federal friction.

Yet, the fundamental vulnerabilities highlighted by the Mythos model have not disappeared. By choosing to let the market dictate the pace of deployment, the federal government has placed an immense bet on the responsibility of private corporations. If an autonomous, highly advanced model eventually breaches critical infrastructure or destabilizes global financial networks, the pressure for reactive, heavy-handed government intervention will return with a vengeance. For now, the administration has made its choice: the race for AI supremacy must be won at all costs, even if it means running toward the finish line with no safety net beneath us.