AI safety fears stalk Trump-Xi summit agenda

Por Phelim Kine and Brendan Bordelon — POLITICO – TOP Stories

The potential dangers of artificial intelligence will hang over Chinese leader Xi Jinping’s visit to Washington on Thursday — whether President Donald Trump likes it or not.

Trump has rubbished those fears as part of a “hoax” aimed at surrendering AI dominance to Beijing. But lawmakers and foreign leaders are piling on the pressure — spurred by rising public concern over the technology — to push Trump and Xi to talk about AI safety.

And yet it may not be enough to move the needle. Both the tech industry and China hawks believe meaningful progress toward a U.S.-China deal on AI safety is unlikely to emerge from the meeting.

“There’s no consensus in industry on what should be achieved in this discussion,” said Paul Lekas, head of global public policy and government affairs at the Software & Information Industry Association, a Washington-based tech trade association whose members include Anthropic, Meta and Amazon. “And I think it’s important to manage expectations and recognize that it will be incredibly challenging to achieve something concrete — let alone binding — in the limited time that’s available here.”

The U.S. and China remain locked in a race for AI supremacy, with the stakes only escalating as the tech grows more powerful. American tech companies accuse their Chinese counterparts of conducting “distillation attacks” on their AI models, allowing them to extract new capabilities and train Chinese models at a fraction of the cost. Meanwhile, Capitol Hill is considering bipartisan legislation that would further choke off supplies of advanced chips to Chinese AI firms.

“There have been a number of points of tension between both countries, and certainly allegations going in both directions,” Lekas said.

Even Anthropic — the safety-conscious AI giant whose CEO Dario Amodei earlier this month called for a coordinated AI slowdown with China — is tempering its expectations ahead of the Xi-Trump summit.

“I don’t think that you’re going to see a deal,” Sarah Heck, Anthropic’s head of policy, said at a POLITICO event last week. “But the fact that we’re starting to talk about these things is what’s most important. There’s probably not a government in the world that thinks it would be advantageous to have some of these risks out there.”

The Chinese embassy didn’t respond to a request for comment. The White House referred POLITICO to comments by a senior administration official, who told reporters last week on condition of anonymity that the topic of AI will be on the summit agenda.

Even as the AI industry tamps down expectations of a sweeping deal on AI safety, its most powerful players are increasingly urging international action.

OpenAI CEO Sam Altman and Anthropic CEO Amodei both addressed the United Nations’ Security Council in New York Wednesday, urging better oversight over AI development to prevent what Amodei called “a risk to humanity.”

That risk may not be theoretical. Earlier this year, the Pentagon narrowly aborted a military operation against a Chinese ship suspected of carrying nuclear weapon components to Iran after officials discovered an AI chatbot had mistakenly identified the ship’s cargo, CNN reported last week. Most Americans worry the technology could someday destroy humanity, per the results of a POLITICO survey published last week.

Escalating fears about the catastrophic impact of AI could push the Trump administration to proclaim a grand bargain with China — regardless of whether a substantive agreement materializes.

“Over the past week, we’ve reached a critical mass of public opinion in the U.S. where it’s clear the Trump administration will feel some pressure to point to discussions with China about AI as a key outcome,” said Ryan Fedasiuk, former adviser on U.S. China bilateral affairs at the State Department in the Biden administration who specializes in the analysis of AI security risks at the American Enterprise Institute. He said that Beijing “has its own rationale for wanting to ensure AI doesn’t inadvertently improve the capabilities of would-be terrorists or cyber-criminals.”

Australia’s prime minister Anthony Albanese urged Trump and Xi to strike a deal on AI safety “in the interests of humanity” in an interview with Australia’s ABC News last week. And in a speech on Tuesday, United Nations’ Secretary General António Guterres called for a “multilateral AI risk management framework.”

The administration has attempted to respond to the growing chorus with vague pledges of bilateral safeguards. On Sunday, Treasury Secretary Scott Bessent proposed a new AI safety “notification mechanism” with Chinese Vice Premier He Lifeng — a direct line between the two officials that is already being downplayed as “BS” by people familiar with the plan.

The White House will launch a new “AI Force” and appoint an “AI Czar” to ensure AI safety, Trump said in a Truth Social post last week. And White House officials — likely sensing a political vulnerability ahead of midterms in November — are quietly organizing a meeting of senior executives of AI companies to discuss the technology’s risks on the sidelines of the summit, CNN reported last week.

Trump doesn’t want to do anything that would “undercut the AI boom in the U.S. and American AI dominance,” said Kyle Chan, who researches China’s technological development at the Brookings Institution think tank. Chan expects Trump to instead “throw in a line about how the U.S. and China should work together on AI — something very generic and abstract.”.

Democratic lawmakers are demanding more. A group of 17 Democratic senators sent a letter to Trump on Wednesday urging him to strike a deal with Xi that would slow or pause AI development in both countries. Rep. Ro Khanna (D-Calif.), ranking member of the House Select Committee on China, called for a “China-U.S. AI treaty” on Wednesday that would curb the race toward AI “superintelligence” and create an independent audit system to confirm compliance.

But most leading Republicans — including House Speaker Mike Johnson — echo Trump’s view that restricting AI development will only benefit China.

“China will break promises on AI,” Rep. John Moolenaar (R-Mich.), chair of the House Select Committee on China, said in a statement Wednesday. The route to AI safety “is implementing strict export controls on chips and chipmaking equipment to continue slowing China’s progress.”

There are myriad obstacles to establishing an agreement between the U.S. and China on AI, and creating a reliable mechanism allowing the two sides to monitor the other’s development and deployment of new systems would not be easy.

“The verification hurdle is pretty high — how do you even quantify the inspections or management of AI driven capabilities?” said Ret. Gen. Leslie Beavers, former director of intelligence, surveillance & reconnaissance enterprise capabilities at the Department of Defense. “You can’t do that the way we count missiles, so how would you impose any kind of traditional safety inspections on the process?

Beijing may demand a high price — including relaxing the existingU.S. export controls on advanced semiconductors — in exchange for concessions on greater transparency into its AI ecosystem.

“The Chinese are going to push us to be more open with our chips, our hardware and our technology,” said Beavers.

But even the modest summit outcome — like the creation of a U.S.-China working group dedicated to AI safety — may help boost transparency on AI development.

“The United States and China have plenty of channels of communication, but they are political — and due to Chinese preferences, none exists yet for technical conversations on AI,” said Mark Beall, who directed AI strategy at the Pentagon during the first Trump administration. “We need the ability to deconflict, deescalate, and share appropriate information in the event of a loss of control incident.”

Fonte: POLITICO – TOP Stories