Retired general credits years of quiet dialogue for 2024 nuclear AI accord

NPR reported that the United States and China are locked in a cycle of fierce competition and mistrust over artificial intelligence, and experts say the chances that President Trump and Chinese leader Xi Jinping can change the dynamic when they meet in Washington next Thursday are dim.

In the past few weeks, the Trump administration has accused Chinese AI companies of stealing intellectual property by “distilling” leading U.S. AI models and threatened sanctions. Beijing, for its part, has poured scorn on U.S. restrictions on chip exports and accused Washington of running a “Cold War playbook” to hold Chinese AI down. It would seem there is little room, or appetite, for compromise at the official level, but former officials and policy experts say there is a buzzing bilateral dialogue — and a growing sense of shared concern about the dangers of AI — at the sub-national level in what diplomacy wonks call “Track Two.”

For decades, government-to-government interactions, sometimes called “Track One” diplomacy, have been underpinned by this sprawling network of informal private meetings and conferences, often run by think tanks or universities. Track Two discussions bring together experts from both sides and cover a wide array of topics, from nuclear disarmament to climate change. The idea behind Track Two is to build understanding and work toward policy recommendations that both sides can agree on. Sometimes, governments use them as back channels or low-stakes ways to float ideas, and Beran said that was the case after direct military-to-military discussions between China and the U.S. were cut off following then-Speaker of the House Nancy Pelosi’s visit to Taiwan in 2022. At the time, Beran was the top China official on the National Security Council in the Biden administration.

“I think there is no more important topic that the U.S. and China should be talking about than AI,” said Sarah Beran, a former U.S. diplomat who worked on China and the incoming president of the nonprofit National Committee on U.S.-China Relations. The committee runs several Track Two discussions and has been involved in policy exchanges since 1972, when it sponsored a visit to China by U.S. ping pong players to help thaw relations.

“The fear of chokepoints, being held back, and AI being used as a cudgel on both sides means that this is also one of the most difficult areas to talk about,” Beran said.

“We floated some ideas in Track Two dialogues about what resumption of military-to-military communications could look like, at what level, what the topics would be,” Beran said of the post-Pelosi-visit period. “And then we took it back into the formal channels and negotiated on resuming those channels.” Then-President Joe Biden and Xi agreed at a 2023 summit in Woodside, California, to get the military dialogue back up and running.

Jack Shanahan, a retired Air Force general who has been involved in Track Two dialogue over military AI, said informal discussions helped tee up an agreement between the leaders a year later that any decision to use nuclear weapons should be controlled by humans, not by AI.

“That was only possible by years of quiet dialogue at the Track Two level on this very topic,” he said.

Samm Sacks, an expert in Chinese tech at the Johns Hopkins School of Advanced International Studies and the policy think tank New America, said she was involved in several Track Two discussions on AI in Shanghai and Beijing over the summer “that were very substantive.”

“People can talk about sensitive issues more candidly and also don’t feel the need to come to any consensus that would become public,” she said.

Those dialogues have reinforced for her that the U.S. has a lot of common ground with China on AI, despite government friction over chip export controls and accusations that Chinese AI labs are stealing intellectual property. “So what I hear in the Track Twos I’m a part of is a sense of shared vulnerability and shared precarity created by AI,” she said.

Shanahan said he has perceived a shift on the Chinese side as AI advances and risks become clearer. “For the first time, I’m seeing, like, ‘Look, we know we’re not at parity in some of these areas, but AI is too important not to have these dialogues,’” he said.

Still, the two sides have tended to focus on different concerns, with the U.S. looking at things like cyber vulnerabilities and possible loss of control of AI, and China thinking about the political stability of its governing system and self-reliance for making hardware, according to experts. That focus has had an impact on the complexion of Track Two dialogues and has tempered expectations for concrete outcomes from next week’s summit, the experts said.

The Center for Strategic and International Studies, a U.S. policy think tank, also runs Track Two discussions, and China specialist Scott Kennedy has been involved in a wide-ranging economics dialogue that includes AI. He said the talks can increase awareness in policy circles of the others’ red lines and how each side makes decisions on the issues.

“They provide you a sort of road map of the different points of view and the potential actions and reactions of different scenarios,” he said. But Kennedy added that the dialogues require government support on both sides to work.

“This is the time where we need some leadership, and what we’re seeing is a vacuum,” Kennedy said. “Track Twos don’t fill that vacuum. They can help redirect things and buy time, prepare the groundwork for new engagements, but they’re not going to solve the giant problems on their own.”

Sacks said the disagreements over AI make this a precarious moment, with the utility of Track Two meetings in the balance. A push this week by U.S. AI companies for a global agreement to slow the development of cutting-edge AI did not go over well in China, she said.

“I actually think the best thing that the two governments can do is just at a high level signal [that] we both need to come together to address AI safety and cyber risk, bio risk, for example,” she said, referring to concerns that AI could be used for hacking or to develop biological weapons.

Some Track Two participants said the Trump administration has been less engaged with the dialogues than previous administrations, making the work more challenging. But the participants said they will be watching the Trump-Xi meeting in Washington next week for signals that AI is a topic of concern to both sides, and one on which the two sides’ experts can continue to engage in substantive talks.

Beran said that includes the National Committee on U.S.-China Relations, which has its next meeting coming up in the next couple of months. “You need to build confidence on both sides that this is a safe space to talk about issues,” she said. “Our next meeting is coming up in the next couple of months and we’re trying to think about how we use whatever the two presidents say together as a way to start talking about these issues in a Track Two format.”