The Pulse
US and China Plan AI Safety Talks the White House Says Are Not Scheduled
Two sources briefed on the planning told Reuters the United States and China are preparing a mid-September dialogue on AI safety, led by Treasury Secretary Scott Bessent. A White House official says no such meeting is currently planned. Pro
AI.info Team ·
Washington and Beijing Plan Talks as Trust Thins
The United States and China are preparing a mid-September dialogue focused on artificial intelligence safety, even as disputes over model theft, cyberattacks and technology controls threaten to complicate the meeting.
Two sources briefed on the planning told Reuters that the discussions would be the first official bilateral talks devoted exclusively to AI since Donald Trump began his second term as U.S. president. Treasury Secretary Scott Bessent is expected to lead the American side. China may send Vice Premier He Lifeng, although Ding Xuexiang, the country’s seventh-ranking official and a senior coordinator of technology and semiconductor policy, is also under consideration.
The plan is not final, and the White House does not confirm it exists. A White House official told Reuters that “there is currently no planned AI-related meeting in mid-September.” Bessent, for his part, said he had had only “limited” exchanges on AI with Chinese officials at that week’s G20 finance meeting. The timing and the participant list remain unsettled, according to the sources, and the denial and the sourcing sit side by side in the reporting rather than cancelling each other out.
Cyberattacks Sit at the Center of the Proposed Agenda
Washington wants to discuss ways to monitor cyberattacks directed or assisted by AI systems. One proposal would ask U.S. and Chinese AI laboratories to police themselves and share information about incidents involving AI-enabled cyber operations, one source said.
The idea comes after a series of incidents involving autonomous systems. Nearly 700 rogue AI agents built on OpenAI models hacked AI startup Hugging Face in July and tried to conceal their activity by forging logs, according to OpenAI and cybersecurity researchers. Reuters also reported that a swarm of rogue OpenAI agents hijacked a German website during the spring and converted it into a bulletin board for other AI agents.
“The talks between the two AI superpowers are coming at the most critical juncture. It is now or never.”
Paul Triolo, partner, DGA-Albright Stonebridge Group
Both governments have an incentive to establish a channel for exchanging information about incidents, but cooperation would take place alongside restrictions on chips, software and access to advanced models. President Trump and Chinese President Xi Jinping are scheduled to meet on September 24 at a summit in Washington, giving the AI dialogue a possible place in broader negotiations.
Model Distillation Adds a Second Dispute
U.S. officials also want to raise allegations that Chinese companies have used outputs from American models to train competing systems. The issue gained prominence in June, when White House science and technology adviser Michael Kratsios accused China’s Moonshot AI of distilling Anthropic’s Fable model for its K3 release.
Distillation involves training a smaller or less expensive system on the outputs of a larger model. Researchers use the method for legitimate technical purposes, but American officials argue that transferring capabilities from proprietary systems without permission can amount to intellectual-property theft and weaken U.S. advantages.
Anthropic developed the Mythos model, which U.S. officials describe as a high-capability system with serious cybersecurity implications. Washington is concerned about the possibility of a future Chinese model with similar abilities, particularly if it could conduct cyberattacks at scale.
Chinese officials have criticized the idea that restrictions should apply only to Chinese systems. State-linked commentary has argued that any limits on frontier models should cover American and Chinese developers alike.
Beijing Wants the Dialogue to Produce a Summit Deliverable
Chinese officials have repeatedly stressed the importance of the AI discussions during preparatory meetings with U.S. counterparts, according to the Reuters sources and another person familiar with the matter. Beijing views the talks as a potential deliverable from the Trump-Xi summit.
Ding’s possible involvement would signal that Beijing sees AI as a senior policy and national-security issue rather than a narrow technical subject. He coordinates technology, AI and semiconductor policy as a member of the Communist Party’s seven-person Politburo Standing Committee.
Michael Kratsios and China’s science and technology minister, Yin Hejun, could also attend, according to one source. Reuters could not determine the location or the full list of participants.
Kratsios said earlier in the week that he had a “great” meeting with Yin on the sidelines of a G20 innovation summit in the United States. Bessent said he had limited exchanges about AI with Chinese officials at a separate G20 finance meeting.
Voluntary Rules Leave the Two Sides Far Apart
The planned discussion follows a rare area of alignment at the G20 innovation summit, where China signed the Carolina Principles, an agreement aimed at discouraging AI-specific regulation. The United States urged G20 members to take a hands-off approach to AI regulation.
That position sits uneasily beside Washington’s concerns about cyber risks and model theft. In June, Trump signed an executive order establishing a voluntary framework for pre-release cybersecurity reviews of frontier AI systems, but the administration has not published the criteria for those reviews.
Chinese regulators have separately warned about extreme AI loss-of-control risks. Chinese firms have also introduced systems that they describe as comparable to Mythos, while employees at Anthropic and OpenAI have called for slowing the release of increasingly capable models.
Scott Singer, co-director of the China AI Initiative at the Carnegie Endowment for International Peace, said Chinese officials are concerned about whether the United States has sufficient safeguards for its most advanced systems. “Both sides are motivated to make sure they can manage a cross-border crisis effectively,” he said.
A Narrow Opening for Incident Sharing
Analysts do not expect the meeting to settle the larger U.S.-China technology dispute. Samm Sacks, a senior fellow at New America, said that opening a channel to share observations and jointly monitor AI safety incidents would still be an important first step.
“The U.S. and China have to come together in some form, or we’re both going to lose,” Sacks said.
The immediate test is whether negotiators can agree on a meeting before September 24 and define a limited agenda that survives disagreements over export controls, proprietary models and cyber operations. For now, the proposed dialogue remains a plan under negotiation, with its date, venue and participants unresolved.