Credit by google
President Donald Trump’s latest remarks downplay AI safety risks, calling the warnings from “negative forces” exaggerated, while stressing that the U.S. must stay ahead of China in the AI race. Meanwhile, several leading AI researchers have recently issued urgent warnings, calling for a slowdown in development. This debate over AI safety and regulation is rapidly heating up across U.S. politics, the tech industry, and the public.
On September 13, 2026, while speaking with reporters in Ireland, Trump directly addressed the question of AI safety. When asked whether the AI industry should slow down or accept more regulation, Trump said some “guardrails” could be put in place, but argued that much of the discussion about AI risk comes from “very negative forces” who are “raising things that won’t happen at all.”
Trump’s core concern is not AI risk itself, but geopolitical competition. He stated clearly: “We lead China in AI. We’re the most advanced nation in the world, and frankly, I want to keep it that way, because whoever wins AI wins everything.”
These remarks came just days after a series of concentrated warnings about AI safety risks from AI industry leaders and researchers.
In early September 2026, the AI safety debate escalated sharply, triggered by a series of researcher departures and public warnings.
Jacob Coxon’s resignation ignited public debate. The 27-year-old researcher, who had worked at both OpenAI and Anthropic, announced on September 9 that he was resigning from Anthropic, just two months before his equity was set to vest. He stated publicly that he was leaving because neither company was acting responsibly, saying they are “racing toward self-improving superintelligence, gambling with our lives.” In an interview, Coxon said AI “could kill all of us before the end of this decade.”
Top AI CEOs spoke out collectively. On September 12, Anthropic CEO Dario Amodei published a 3,800-word essay calling for a global slowdown in AI development. He wrote: “Over the past few months, I have become increasingly convinced that adequately addressing the risks requires more prudence — not just investing in risk prevention, but controlling the pace of capability advancement so that risk prevention has time to catch up.” He proposed introducing independent third-party evaluators to verify companies’ safety practices.
OpenAI CEO Sam Altman responded on social media hours later, saying a slowdown is necessary and that “committing to give independent evaluators employee-like access is a good idea.” Google DeepMind Chairman Demis Hassabis also said Amodei’s direction “is correct.” Elon Musk’s response was more concise: “Dario is right.”
Broader industry concern. In recent years, Geoffrey Hinton resigned from Google to speak freely about AI risks, and Turing Award winner Yoshua Bengio has consistently pushed for stricter regulation, advocating a pause on training advanced models until safety parameters are established. Executives including Google DeepMind’s Demis Hassabis, OpenAI’s Jakub Pachocki and Sam Altman, and Anthropic’s Dario Amodei have jointly committed to treating the mitigation of AI extinction risk as a global priority.
The focus of this debate is not whether AI carries risk, but the nature of that risk, its urgency, and how to respond.
One side advocates cautious advancement. Experts like Amodei argue that AI capabilities are advancing faster than safety research can keep up. Amodei specifically mentioned “recursive self-improvement” — AI systems becoming more advanced on their own without human researchers. He warned: “If left unchecked, it could exceed our ability to understand and control these systems.”
The other side fears regulation becomes a competitive barrier. Trump and his supporters’ core argument is that overregulation could leave the U.S. behind in its race with China. David Sacks, co-chair of the White House Science and Technology Advisory Council, criticized on social media that if government agencies and a handful of AI companies jointly hold the “gatekeeping power” over model releases, it could concentrate control over AI development in a few hands. He argued that Anthropic’s push for regulation may aim to restrict open-source models, with the result potentially being that “while other countries have more choices, the U.S. becomes an island full of expensive closed-source models.”
This concern is not unfounded. Reporting by Axios noted that faced with a potential regulatory system, OpenAI, Google, and Anthropic all have the lawyers, safety teams, and government relations staff needed to navigate complex certification regimes, giving them a clear advantage over startups and open-model developers. Critics worry that safety-focused rules could become a tool for a few major companies to reinforce their own advantages.
Trump repeatedly stresses that “we lead China in AI,” but the latest research data presents a more complex picture.
The U.S. still has an edge in top models, but the gap is narrowing. Stanford University’s 2026 AI Index Report shows the U.S.-China AI model performance gap has “substantively closed.” Since early 2025, leading systems from both countries have repeatedly traded the top spot. By March 2026, Anthropic’s top model led by only 2.7%. U.S. Treasury Secretary Scott Bessent previously acknowledged that America’s AI lead may be only three to six months.
China leads in scale, patents, and applications. In 2024, China accounted for 74.2% of global AI patent grants (97,206 out of 131,121), while the U.S. share fell from 42.8% in 2015 to 12.1%. In academic output, China contributed 17.8% of global AI research papers, more than double the U.S. share of 7.6%. Among the world’s top 100 AI companies, China has 51 and the U.S. has 37.
The two countries’ AI ecosystems are structurally different. The report describes the relationship as “complementary competition”: China has a numerical advantage in large models and application deployment, with strong commercialization capabilities; the U.S. maintains leadership in the framework layer and high-end chips, and dominates the open-source ecosystem and core development tools.
The Trump administration’s AI policy shows a clear deregulatory bent, though it retains some caution in national security areas.
Rescinding Biden-era AI safety order. On his first day in office in January 2025, Trump rescinded Biden’s 2023 AI safety executive order (EO 14110). He subsequently issued several executive orders, including “Removing Barriers to American AI Leadership” in January 2025, the “American AI Action Plan” in July, “Project Genesis” in November, and the “Ensuring a National Policy Framework for AI” in December.
March 2026 national AI policy framework. This framework contains seven pillars, with a core position of opposing the creation of a new federal AI regulatory agency by Congress, advocating that existing sector regulators handle AI applications in their domains, and pushing for federal “preemption” of state-level AI laws.
June executive order focused on cybersecurity. The “Promoting Advanced AI Innovation and Safety” executive order signed on June 2, 2026, established a voluntary safety deployment framework for “covered frontier models.” AI developers may voluntarily give the government up to 30 days of pre-release access for safety evaluation. The order explicitly states it does not authorize any mandatory government licensing or pre-approval requirements.
Key gap: mandatory safety guardrails for frontier models. Analysis by NYU Stern’s Center for Business and Human Rights points out that the “deregulatory core” of Trump’s framework is concentrated precisely in the most critical area — frontier AI development itself. The framework requires the federal government not to intervene in this area or create a new regulator, relying only on “industry-led standards.” The analysis warns: “The framework does not answer what may be the most important question: what happens when the next safety commitment quietly disappears. That silence is itself a policy choice — and a consequential one.”
The debate over AI safety will not subside anytime soon, and several key variables will determine the direction.
Whether Congress intervenes. Currently, the U.S. has no comprehensive federal AI legislation. Trump’s framework asks Congress to legislate on child safety, AI fraud, intellectual property, and workforce training, but explicitly opposes creating a frontier AI regulator. If Congress continues to refrain from acting, state-level laws (such as Colorado’s AI Act and California’s AI transparency law) will remain in effect, but face legal risk from federal “preemption” challenges.
The credibility of industry self-regulation. The independent third-party evaluations Amodei calls for and the “employee-level access” evaluators Altman has promised, if implemented, could become de facto industry standards. But if safety commitments fade under competitive pressure — as NYU Stern’s analysis notes, Anthropic has revised its “responsible scaling policy” and OpenAI has disbanded its alignment team — public trust will erode further.
The direction of public opinion. Multiple polls show rising public concern about AI in the U.S. High-profile researcher departures and warnings are pushing technological risk from academic discussion onto the public agenda. Trump’s remarks about AI regulation in a Fox News interview — including claiming he “shut down” Anthropic (the actual context was a dispute over the company’s access to the Pentagon) — have sparked widespread criticism on social media. This public pressure could eventually force policy changes, even if the White House currently leans toward a hands-off approach.
The essence of this debate is an unresolved governance dilemma: how to maintain technological leadership without letting “leadership” itself become an excuse to ignore risk. The tension between Trump’s “whoever wins AI wins everything” and Amodei’s “we must slow down” will define the next phase of U.S. AI policy.