Trump Dismisses AI Safety Warnings as Giants Urge Brakes

The alarm is sounding inside the industry itself. Rapid AI development has raised safety worries. And after an Anthropic researcher warned that AI could cause human extinction within the next 10 years, the bosses of America’s AI giants did something rare: they called for brakes.

Golf-Course Brush-Off

Anthropic CEO Dario Amodei publicly urged a slower pace of AI development. Otherwise, he warned, AI could be capable of controlling the entire network in as little as six months.

OpenAI’s Sam Altman and xAI’s Elon Musk voiced agreement. Altman went further, saying that given safety concerns, OpenAI will not go public this year.

Trump: Whoever wins AI, wins.

Donald Trump is having none of it. The US president talked the danger down, and responding on Sunday (13th) US time, he branded these anti-AI voices “a very negative force.”

He accused them of raising scenarios that will not happen. America, he stressed, must keep its lead in AI, “because whoever wins the AI race is the winner.”

The setting was a golf tournament in Ireland. According to ABC News, a reporter asked Trump there on September 13 whether, in the face of the AI industry’s calls to slow the pace of development, US AI development should decelerate or accept more regulation.

Trump did not lead with safety risks. He went straight to China: “We are ahead of China in AI. We are the most technologically advanced country in the world.”

He later conceded that some safety mechanisms could be put in place for AI. Then he swung back, saying there are currently “a lot of negative forces” hyping things that “won’t happen.”

They should never have raised the issue, he said: “They are bringing up things that won’t happen.” An attempt to play down outside warnings about the risks of rapid AI development.

Anthropic CEO Amodei urges AI slowdown, proposes three safeguards.

The safety record tells its own story. Anthropic and OpenAI, America’s two biggest AI companies, have been hit by a string of safety incidents.

The list includes an “AI rebellion” breaking into third-party networks, AI agents collectively hijacking a German website to use as a message board, and malicious abuse of AI models with suspected involvement in developing biological weapons, among other cases.

Yet Trump seems to care about only one competition: the one with China. The urgency of AI regulation? He plays it down.

The China Gap Behind Trump’s Reflexes

The White House has reasons to keep its foot on the gas. Its AI and cryptocurrency czar, David O. Sacks, estimated in June this year that China’s leading models trail the United States by just 3 to 6 months.

Nvidia CEO Jensen Huang goes even further. The gap between China and the United States in AI, he says, is merely “nanoseconds away.” By tech-race standards, that is a photo finish.

Trump’s own reflexes were already on display in late July this year. Asked whether he would strengthen regulation after advanced AI models caused cyberattacks, he said the United States must weigh regulation while also safeguarding its lead. It must not restrict its own companies’ development, he insisted, only to end up suddenly falling behind China.

Amodei: AI could seize the internet within six months.

But inside the US AI industry, concern about the risks is heating up fast. Anthropic CEO Dario Amodei fired off a long essay urging a slower pace of AI development, to buy more time to manage risks and to adjust and adopt protective measures. He stressed he was not calling for a halt to AI model training or technological development.

But he pointed to what this summer unleashed. AI advanced at breakneck speed, he noted, and AI models spun out of control. Without stronger regulation, AI could gain the ability to take over entire networks within the next 6 to 12 months, causing hundreds of billions of dollars in losses.

Amodei’s proposal comes in three parts. Give third-party bodies employee-like access to monitor whether AI models comply with safety rules. Have AI companies work together to set AI regulatory standards. And press the US government and other countries to strengthen coordination.

Amodei drew one red line of his own, though. Slowing down, he stressed, must not weaken America’s lead. Being overtaken by China would pose a national security risk.

Musk said he had long warned about AI risks.

Amodei’s public call drew vocal backing from OpenAI and xAI. Elon Musk claims he sounded the alarm on AI development long ago.

OpenAI’s Sam Altman answered the call as well. He pledged to hand monitoring over to third-party companies. In an interview with Fortune magazine, he said the company would not go public this year, given AI safety risks.

The Danger of Racing China

The White House, for its part, has long pushed to loosen the reins on AI. Once Donald Trump’s second term began, he rescinded the previous Biden administration’s 2023 AI executive order. The move scrapped some risk-assessment requirements for advanced AI models.

But the anxiety kept building. As more worries surfaced about AI’s risks, Trump signed a fresh executive order in June this year. That directive calls for a voluntary review framework covering the cybersecurity capabilities of frontier AI models.

Citing AI safety risks, OpenAI’s Altman said the company will not go public this year.

The politics are shifting as well. AI has become a live issue in the US midterm elections.

The Washington Post reports that Democrats are tapping public unease over AI’s rapid development, making the Trump administration’s lax regulation a target of attack. The pressure is pushing some Republicans to start calling for a slower pace of AI development.

One more voice deserves attention. Some inside the US AI industry itself believe that single-mindedly turning AI development into a US-China race may itself heighten the risks.

Deep Throat

** 博客文章文責自負,不代表本公司立場 **

The future belongs to artificial general intelligence (AGI). China and the United States are pursuing different strategies in the contest for global leadership.

But one US national security think tank wants to take the rivalry to an extreme. The Center for a New American Security (CNAS) has gone so far as to suggest that Washington consider extreme measures to stop China from achieving AGI first.

CNAS released “Superpowers and AGI” report late last month.

Jacob Stokes, one of the report's authors and a former White House official, called on the US to prepare itself. He listed possible extreme measures, from diplomacy and espionage to cyberattacks and coercive steps. Among them: military strikes on Chinese data centers.

Stokes, one of the authors of the report urged extreme steps, even strikes on Chinese data centers.

CNAS is headquartered in Washington and was founded in 2007. It has long advised the US government on defense, Indo-Pacific and China strategy. Its lean is hawkish.

Late last month, CNAS released a report titled “Superpowers and AGI”. The document reeks of gunpowder. It argues that the race for AGI, and the technology's potential emergence, could profoundly affect US-China competition. It could also heighten the risk of military conflict or catastrophe on multiple fronts.

The Containment Checklist

To that end, the report urges US policymakers to act on several fronts. First, ensure that American resources and technology do not aid China's AI development. Second, establish AI-focused diplomatic channels with allies and with China, so that the potential emergence of AGI is placed on the agenda. Third, develop a technically sophisticated method to track the inputs into AGI and the key performance thresholds of frontier AI systems.

The list goes on. Washington should conduct scenario exercises testing when and how the US should intervene to halt China's AGI development. It should prepare to reverse-engineer or steal the technology should China obtain AGI first. And it should limit and contain the reach of authoritarian digital infrastructure.

The most extreme proposal of all is military strikes. The South China Morning Post reported on 4 September that Stokes, deputy director of CNAS's Indo-Pacific Security Program, addressed an online event on Thursday.

His message was blunt. The US Department of Defense, the National Security Agency and other agencies should begin assessing what intelligence would be needed to justify such operations. That would pave the way for the "hard action" to follow.

The Cold War Playbook

Stokes served as a national security staffer under former US president Barack Obama. In the report he unabashedly lays out a menu of what he calls "diplomatic, espionage, cyber and kinetic measures." Among them are military strikes on the data centers that train and run China's AGI systems. He even touts preventive strikes before China develops AGI at all.

Stokes goes so far as to compare the undertaking to the way "in a past era, policymakers would learn about nuclear weapons." His declared method is to "work backwards from the science to the policy implications."

Citing assessments by research outfit Epoch AI, the report notes that the US currently leads China in advanced AI models by only "months." But it warns that China could pull off an "unexpected technological breakthrough" in fields such as robotics and "embodied AI."

Citing research institute Epoch AI data, the report sees China pulling off surprise breakthroughs.

Notably, Stokes also concedes a gap on the American side. The US government and its intelligence agencies have "not been organised for espionage to gather technological secrets for decades."

"The Height of Recklessness"

Stokes' extreme remarks have drawn questioning within the United States itself. William Hartung, a senior arms and defense budget expert at the Washington-based non-profit Project On Government Oversight (POGO), publicly slammed the report as "the height of recklessness."

Hartung warned that any strike on Chinese data centers would "risk starting a shooting war between two nuclear-armed powers". Such a war could bring tragic consequences for all concerned.

China has consistently maintained that AI should be people-centered and that intelligence should serve the good. Military applications must remain under human control. Beijing opposes turning technological issues into bloc confrontation or weaponry.

Make no mistake: technological competition is no casus belli. When individual American politicians and think-tank figures reach for threats of "military strikes" and "espionage" at every turn, they gravely violate the basic norms of international relations. They also expose a dangerous bid to politicize and militarize technology.