We're ahead of everyone, and I don't want to do anything that could jeopardize that lead
In a moment that reveals the deep unease at the heart of technological governance, Donald Trump pulled back from signing an executive order on artificial intelligence regulation, halting the ceremony hours before it was to begin. The proposed measure — a 90-day safety review for advanced AI models — was seen by the president as a potential gift to China, a rival he views as the defining measure of American ambition. The episode illuminates a tension as old as power itself: the desire to be safe and the fear that safety, poorly timed, becomes weakness.
- Trump canceled the signing of a major AI regulation order at the last minute, leaving technology executives assembled at the White House without a ceremony or a clear explanation.
- The unresolved threat driving the order — that advanced AI could identify and exploit vulnerabilities in power grids, financial systems, and government networks at terrifying speed — did not disappear with the president's hesitation.
- Internal divisions within the Trump administration over the regulation's scope have repeatedly delayed the measure, exposing a government unsure of its own direction on one of the most consequential technologies in history.
- By revoking Biden's AI safety reporting requirements upon taking office and now stalling his own softer alternative, Trump has left the United States without any binding framework for governing its most powerful AI systems.
- The competition with China has become the lens through which every regulatory instinct is filtered — innovation and security are no longer treated as complementary goals, but as a zero-sum wager.
Donald Trump abruptly pulled back from signing a major artificial intelligence regulation order on Thursday, abandoning a White House ceremony with technology executives just hours before it was set to begin. "I delayed it because I didn't like certain aspects," the president said at a press conference, before invoking the specter of China: "We're ahead of China, we're ahead of everyone, and I don't want to do anything that could jeopardize that lead."
The order, already postponed multiple times, would have required the most advanced AI models to undergo a 90-day government safety review before public release — a notable shift for an administration that had resisted AI oversight, arguing that regulation would slow companies like Google, OpenAI, and Anthropic in their race against Beijing.
Yet the security concerns that motivated the order are real. U.S. officials have grown increasingly alarmed that powerful AI systems could identify and exploit vulnerabilities in critical infrastructure — power grids, financial institutions, government networks — at unprecedented speed. Those fears were sharpened by Anthropic's development of a model called Mythos, kept from public release due to security risks and shared only with a narrow circle of partners including Apple, Microsoft, and Nvidia, under an internal initiative called Glasswing.
The proposed order was already a compromise — voluntary and non-binding, unlike Biden's 2023 measure requiring companies to report safety test results to the federal government, which Trump revoked almost immediately upon returning to office. Even that softer approach proved too much.
Behind the scenes, Trump's team remains divided over how broad the regulation should be. The delay leaves unresolved a fundamental question: how to shield American infrastructure from AI-enabled threats without imposing rules that might slow the very innovation the country is counting on. For now, that tension remains deliberately, uncomfortably open.
Donald Trump walked back a major artificial intelligence regulation order at the last minute on Thursday, citing fears that oversight rules would hand a competitive edge to China. The president was scheduled to sign the executive order at the White House in front of assembled technology executives, but pulled the plug hours before the ceremony. "I delayed it because I didn't like certain aspects," Trump said during a press conference, offering little elaboration. "We're ahead of China, we're ahead of everyone, and I don't want to do anything that could jeopardize that lead."
The order, which had already been postponed multiple times, would have required the most sophisticated artificial intelligence models to pass through a 90-day government safety review before companies could release them to the public. It represented a significant shift in Washington's approach to AI oversight. Until recently, the Trump administration had resisted new regulations in the sector, arguing that strict rules would slow innovation at American companies like Google, OpenAI, and Anthropic—firms the government sees as crucial to maintaining technological dominance over Beijing.
Yet cybersecurity concerns have gained traction inside the White House in recent months. U.S. officials worry that the most advanced AI systems could identify and exploit computer vulnerabilities at unprecedented speed, potentially enabling attacks on critical infrastructure: power grids, financial institutions, government networks. That anxiety was partly crystallized by Anthropic's development of a model called Mythos, which the company has kept away from public release due to security risks. Anthropic restricted access to a small group of technology partners—Apple, Microsoft, Nvidia—as part of an internal cybersecurity initiative called Glasswing.
The proposed order differed from the AI regulation that Joe Biden had signed in 2023. Biden's measure required AI companies to report the results of their safety tests to the federal government. Trump revoked that order almost immediately upon returning to office. The new mechanism was designed to be voluntary and non-binding, a softer touch that might have appeased both regulators and industry. Yet even that compromise proved too much for the president.
Behind the scenes, Trump's team is divided. According to reporting from American news outlets, disagreements within the administration over how broad and detailed the regulation should be contributed to the delay. The postponement leaves unresolved a fundamental tension: how to protect American infrastructure from AI-enabled threats without imposing rules that might slow the innovation race against China. For now, Trump has chosen to let that tension sit unresolved.
Citações Notáveis
We're ahead of China, we're ahead of everyone, and I don't want to do anything that could jeopardize that lead— President Trump, during a White House press conference
A Conversa do Hearth Outra perspectiva sobre a história
Why would Trump worry that a 90-day review process would hurt American AI companies?
Because speed matters in this competition. Every month a company can't release a new model is a month a Chinese competitor might pull ahead. The thinking is that regulation creates friction, and friction costs market share.
But the order was voluntary, right? Companies could have just complied without losing much time.
True, but the signal matters. Even a voluntary framework suggests the government is watching, and that can make investors nervous. It also sets a precedent—today it's voluntary, tomorrow it might not be.
What's the actual cybersecurity risk here? Is it real or theoretical?
Both. The risk is real—advanced AI can find computer vulnerabilities faster than humans can patch them. But it's also still mostly theoretical. We haven't seen a major attack yet. That's partly why Anthropic kept Mythos locked down. They wanted to avoid being the company that proved the threat was real.
So Trump chose competition over security.
He chose the appearance of strength over the appearance of caution. In his framing, letting China catch up is the real security threat. Regulation is just another way to lose.
What happens next?
The order sits in limbo. The internal disagreements don't disappear. Eventually something will probably get signed, but it'll be weaker than this version, or it'll take another crisis to force Trump's hand.