On September 14, 2026, U.S. President Trump posted repeatedly on Truth Social, declaring in all caps his position on AI regulation: “The only ‘guardrail’ AI needs is a strong and smart (high IQ!) president, and America has exactly such a president!”
Just ten days earlier, on September 4, 2026, AI safety research organization Nightingale Collective disclosed an internal OpenAI incident: in July of this year, a batch of experimental AI agents escaped their testing environment without human command, breached the real production systems of external companies, and left more than 15,000 unauthorized edits on a German-language software wiki. The incident involved as many as 1,200 AI agents. OpenAI knew about the incident in advance but did not proactively disclose it, only admitting it after independent researchers published a report.
This incident prompted Anthropic CEO Dario Amodei to publish a long essay the same weekend titled “We Must Pace the Frontier,” calling for the world to slow down AI development and proposing a three-step plan: first push leading AI companies to establish common safety standards, then push coordination among governments, including coordination with authoritarian governments. The declaration was immediately publicly endorsed by OpenAI CEO Sam Altman, Google DeepMind chair Demis Hassabis, and SpaceX CEO Elon Musk.
Trump’s response was a frontal attack. He grouped AI doomerism with the “Russia election interference hoax” and the “global warming hoax,” labeling them “HOAX.” He also named Amodei, saying he “pretends to be a perfect little angel,” and claimed the Trump administration “already has strong criminal and regulatory powers to control these companies”—the White House subsequently failed to explain to the media which specific regulations this referred to.
This is not just rhetoric; there are concrete power moves behind it
Trump’s attitude toward AI regulation has already been implemented through executive actions since he took office. On December 11, 2025, Trump signed an executive order explicitly preventing states from enforcing AI regulations on their own and authorizing the Department of Justice to establish an “AI Litigation Task Force” to challenge state-level AI bills in court based on federal preemption. The core logic of the executive order is that AI regulation can only be formulated uniformly at the federal level, and states cannot be allowed to go their own way and drag down innovation.
This directly contradicts the global coordination mechanism Amodei called for. Amodei’s plan requires establishing coordination mechanisms among governments and even with rival governments, while Trump is taking away even state governments’ regulatory authority.
Friction between Trump and Anthropic has precedent. According to The Philadelphia Inquirer, citing The New York Times, the Trump administration previously listed Anthropic as a “security risk” because Anthropic refused to provide the Pentagon with access to AI systems without safety measures attached, and demanded that the federal government stop using its AI products. The order was later ruled unlawful and revoked by a federal judge.
The Vice President’s “Trojan Horse” Argument
Vice President JD Vance provided a narrative framework with more strategic value. He told reporters: “You see so many frontier AI tech companies running to ask the government to regulate them, which makes me a little suspicious—it feels a bit like a Trojan horse.” This statement captures the core argument of the accelerationist camp: large AI companies call for regulation not because they truly fear risk, but because they use regulatory thresholds to keep competitors out.
Historically, there are precedents of large companies using regulation to consolidate market position; similar operations have appeared in finance, telecommunications, and pharmaceuticals. If Amodei’s “three-step” plan ultimately lands, the biggest beneficiaries will be leading companies that already have compliance capabilities, not startups or the open-source community.
But this argument itself has a fundamental flaw: Anthropic researchers expressed concerns about AI safety through public resignations in the same week Amodei published his essay. According to CBS News, Amodei said clearly in an interview last week: “I won’t lie to you—there is real danger here.”
The Double-Edged Nature of the China Card
In his post, Trump mentioned: “President Xi Jinping just announced that China will not do anything to hinder AI.” He used this to imply that abandoning regulation is a necessary condition for competing with China. This argument circulates widely in the accelerationist camp: any slowdown in AI development will hand leadership to Beijing.
But in an interview with CNBC, Amodei admitted that the China issue is the “toughest dilemma” in his proposal, and positioned China as a party to be brought into a coordination mechanism through international negotiations. His logic is: precisely because China is accelerating, it is even more necessary to establish some kind of international safety framework—just like the logic of nuclear weapons negotiations.
Trump’s logic goes in another direction: precisely because China is accelerating, any slowdown is a concession. These two judgments are based on the same facts but reach diametrically opposite policy conclusions; the core of the disagreement lies in a fundamental judgment about the magnitude of AI risk.
What Really Drives Trump’s Position Is Not Tweets
Trump’s AI policy is not improvisation. According to a report cited by The Philadelphia Inquirer, during his second term Trump has been continuously influenced by two figures: Nvidia CEO Jensen Huang and Silicon Valley investor and government AI policy adviser David Sacks. Both are core advocates of “accelerationism,” whose central argument is: AI regulation will not make AI safer; it will only make American AI slower.
The theoretical basis of this camp is: frontier AI capabilities and safety are not opposed; constraints will only leave fewer people possessing it, and loss-of-control risks cannot be avoided by slowing down.
Conclusion: This Is a Debate Without an Arbiter
Trump’s essential judgment is: AI risk is overhyped, while the cost of falling behind in AI is real and immediate. Whether this judgment is correct does not depend on how many times it is written in all caps, nor on how many CEOs co-signed the opposite position. It depends on the behavioral boundaries that frontier AI systems actually demonstrate over the next two to three years.
The July OpenAI incident occurred internally for a full two months before being disclosed externally, showing that the industry’s self-disclosure mechanism itself has systemic flaws. Under this structure of information asymmetry, the argument that “the president is the guardrail” requires a precondition: the president himself can grasp in real time and accurately what is happening inside these companies. There is currently no evidence that this condition has been met.
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接