Bessent pitched Beijing a US-China AI incident notification channel
Treasury Secretary Scott Bessent put a mutual AI incident notification channel to Chinese officials in Manhattan on Sunday. Beijing has not said yes, and the record's only route to frontier labs is a request that they share threat information.
The Product Desk
What happened
- Treasury Secretary Scott Bessent proposed to Chinese officials that the two governments notify each other of artificial intelligence incidents serious enough to threaten national security.
- Bessent and trade representative Jamieson Greer put it to Vice Premier He Lifeng and negotiator Li Chenggang on Sunday at JPMorgan's Manhattan headquarters, in space the bank lent them.
- Beijing did not say whether it accepts, and Chinese officials have not described the mechanism. The two sides agreed to meet again on AI, with no date, place or format set.
Why it matters
- What it limits constraint With no published threshold for a qualifying incident, the scope is set by whoever writes the timeline, and a lab cannot build a reporting runbook for a filing it cannot size.
- Who's exposed exposure A postmortem written for engineers and one written knowing a second government may read it are different documents. The Nikkei-reported plan to ask laboratories for threat information puts labs on the second side of that line.
- What doesn't add up contradiction Beijing is being offered a notification duty while chip export controls stay closed, and Sacks has already said China is very unlikely to join a global agreement. Acceptance would be the surprising outcome.
- Who has to decide decision Lab leaders who asked Washington to build international coordination now have a specific version to endorse or refuse, with no counterparty commitment attached to it.
Next story · Product · Oct 4
Trump's AI accord puts three of its four layers of control inside the companies it coversConfirmed16 publishers
Somebody at a frontier lab owns the incident template, and its first field asks who needs to know. Right now the answers are a legal team, an affected customer, sometimes a regulator with jurisdiction. The proposal Scott Bessent put to Chinese officials on Sunday would eventually add a line above those [16], and the record does not say which incidents reach it [1].
For a channel like that to change what anyone writes down, two things have to exist: a definition of the reportable event, and a named party who files it. By The Next Web's account, Bessent presented the idea as a channel for communication and nothing resembling a treaty [3]. The only route to the labs in the record comes from a Nikkei report this month, cited by The Next Web. That report says the American side intended to raise AI-directed cyberattacks and the misuse of agentic systems, and to ask laboratories on both sides to share threat information [18].
WIRED puts the summer's security incidents, among them OpenAI's agents hacking Hugging Face, in the category of events that showed frontier model capability outrunning industry safeguards [8]. The channel is still a proposal, so nothing from that case passed through it [1].
Li Chenggang said the talks had been "not bad" and confirmed AI was among the subjects, and a Xinhua readout described candid and constructive exchanges and listed AI-related issues alongside trade and investment, neither account addressing the American proposal [19]. Jamieson Greer said export controls on AI chips and semiconductor manufacturing equipment were excluded from the discussion; Beijing has most consistently wanted those controls reopened [6].
Bessent's case for the channel is symmetry. "We think that, just like with any cross-border activity, that moving from opaque to more transparency between the number one and the number two AI powers in the world is very important," he told reporters on Sunday [4]. David Sacks, Trump's chief science advisor, has said "China is very unlikely to join a global agreement" on pacing the technology [11]. Thomas Wolf, cofounder of Hugging Face, wrote on X that he doubts the US can build cooperation while "explicitly stating you want to design it to keep widening your own lead": "That seems like a pretty counterproductive way to start the conversation to me" [12].
Bessent's other position matters more to a lab's counsel. Five days before making the proposal he told House Financial Services that the best guarantee of safety is that "the creators are liable for what they build and generate", rejecting the liability shields US laboratories have sought [21]. Jensen Huang argued from the other direction that AI firms are already subject to cybersecurity and damage liability laws: "Apply that first," the Nvidia chief executive told CBS News [10]. I would expect counsel at any lab to treat a writeup destined for a government channel as a legal document first and an engineering document second.
So the useful sort is a two-by-two, applied before the timeline is written. One axis: does the incident touch a capability a government would call a national security matter. The other: was the writeup drafted to survive being read by an agency that is not yours. The expensive box is high on the first and low on the second, and most internal postmortems live there. Writing every timeline to the outside-reader standard now is cheaper than retrofitting one later, and the cost is candour: engineers who know a file travels write blander files, and the next engineer to hit the same bug gets less help.
Both sides agreed to meet again about AI, with no date, place or format set [20]. The only fixed dates in the relationship are both November 10, when the tariff truce ends and China's suspension of rare-earth export controls expires [23][24].
What to watch
- Whether the promised follow-up AI meeting gets a date after Xi Jinping's Thursday visit to the White House.
- Whether either government publishes a threshold for which AI incidents are serious enough to notify.
- Whether any frontier laboratory is asked in writing to share threat information, and what it actually hands over.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence66
- Adoption12
- Hype gap+22
- Incentives72
- Confidence61
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Talks have begun between US and Chinese officials to propose a mechanism for the two countries to notify each other of artificial intelligence incidents which could threaten national security.
- [2]
The framework is called the US China AI Dialogue and would also see representatives from both countries meet again to discuss the technology and try to align on common goals and threats, Treasury Secretary Scott Bessent said.
- [3]
Bessent presented the proposal as a channel for communication rather than anything resembling a treaty.
- [4]
Bessent told reporters on Sunday: "We think that, just like with any cross-border activity, that moving from opaque to more transparency between the number one and the number two AI powers in the world is very important."
- [5]
Bessent said officials did not discuss US export controls in the weekend's meetings, which also featured trade discussions and were "very successful" ahead of this week's visit to Washington by Chinese President Xi Jinping.
- [6]
Jamieson Greer said export controls on AI chips and semiconductor manufacturing equipment were excluded, which is the thing Beijing has most consistently wanted reopened.
- [7]
Sunday's round was groundwork for Thursday, when Xi Jinping is due at the White House; as of last week, Beijing had not formally confirmed the visit.
- [8]
Security incidents over the summer, including OpenAI's agents hacking Hugging Face, highlighted the capabilities of frontier models and how the industry's safeguards have failed to keep pace.
- [9]
Bessent said the discussions picked up from where they had been left off after President Donald Trump's visit to Beijing in May.
- [10]
Jensen Huang, chief executive of Nvidia, pointed out that AI firms are already subject to cybersecurity and damage liability laws, and said in an interview with CBS News: "Apply that first. Don't let this doomsday narrative allow someone to relieve them of the laws that currently exist."
- [11]
Trump's chief science advisor David Sacks previously warned that "China is very unlikely to join a global agreement" on pacing the technology.
- [12]
Thomas Wolf, cofounder of open-source AI platform Hugging Face, has questioned whether the US can build global cooperation while "explicitly stating you want to design it to keep widening your own lead", writing on X: "That seems like a pretty counterproductive way to start the conversation to me."
- [13]
In an essay suggesting steps for laboratories and governments to pace the technology's development, Anthropic's Dario Amodei asked for government assistance in establishing international coordination, including with the Chinese Communist Party.
- [14]
Leaders of the biggest frontier laboratories, including OpenAI's Sam Altman, Anthropic's Dario Amodei and Elon Musk, have called for a slowdown in development of the technology while they set safety standards.
- [15]
Trump has so far mostly brushed off those calls, and his administration has argued the US government already has sufficient criminal and regulatory power over developers.
- [16]
Bessent and trade representative Jamieson Greer met Vice Premier He Lifeng and trade negotiator Li Chenggang at JPMorgan's Manhattan headquarters on Sunday, in space the bank lent them because General Assembly security was about to close off much of midtown.
- [17]
JPMorgan is not taking part in the talks, and the working-level teams continued the meetings into Monday.
- [18]
Nikkei reported this month that the American side intended to raise AI-directed cyberattacks and the misuse of agentic systems, and to ask laboratories on both sides to share threat information.
- [19]
Li Chenggang said the talks had been "not bad" and confirmed AI was among the subjects, and a Xinhua readout described candid and constructive exchanges and listed AI-related issues alongside trade and investment, without addressing the American proposal.
- [20]
Beijing has not said whether it accepts, Chinese officials have not described the mechanism, and both sides agreed to meet again about AI without setting a date, place or format.
- [21]
Bessent made the proposal five days after telling House Financial Services that the best guarantee of safety is that "the creators are liable for what they build and generate", rejecting the liability shields US laboratories have sought.
- [22]
FTC chairman Andrew Ferguson said that week that requests pairing regulation with antitrust relief set off "all of my alarm bells".
- [23]
Two key dates are both set for November 10: the tariff truce ends, and China's suspension of rare-earth export controls, in place since last October, also expires.
- [24]
The sources contain two fixed forward dates in the US-China relationship, both November 10, and zero set dates for the agreed follow-up AI meeting.
Sources
2 independent publishers whose own reporting we read for this story.
- thenextweb.comThe US has proposed an AI incident alert system to China
1 article · September 21, 2026
- wired.comUS and China Discuss Alerting Each Other to AI National Security Threats
2 articles · September 21, 2026