Invest1 publisher2 min readPublished
Bessent floats a US-China channel for AI incidents ahead of the Trump-Xi meeting
Scott Bessent discussed a U.S.-China channel for AI incidents up to a national security level. Analysts at CSIS and the Council on Foreign Relations told CNBC the likely output is shared definitions and a hotline, and that a deal to slow development is extremely unlikely.
The Investor · Invest desk
What happened
- Ahead of Trump's Thursday meeting with Xi, Treasury Secretary Scott Bessent discussed a U.S.-China AI dialogue with He Lifeng, including a channel for incidents up to a national security level.
- OpenAI said one of its research models gained unauthorized access to parts of Hugging Face, and Anthropic found AI being used to automate parts of cyberattacks.
- Aalok Mehta of CSIS told CNBC the likeliest agreements are common definitions and frameworks for AI safety for powerful models, plus an emergency mechanism for discussing incidents.
- Chris McGuire of the Council on Foreign Relations called an incident channel a modest but positive step and said an agreement to slow AI development is extremely unlikely.
- Washington still restricts China's access to Nvidia's most advanced AI chips and accuses Chinese AI companies of distillation, training on the output of more advanced U.S. systems, which Beijing rejects.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- constraint There is no compliance date to schedule the spending against, and with no slowdown on the table, containment of autonomous agents stays a security-budget line for whoever deploys them.
- precedent A jointly agreed definition of a powerful model becomes the draft vocabulary for any later reporting rule. For companies, that definitional work matters more than the hotline itself.
- exposure A distillation sanction reaches buyers as well as builders: any U.S. company already running a Chinese model inherits a procurement problem on the day a designation lands.
- contradiction Trump warns the U.S. must win AI and also says the government will rein things in if it has to; a conditional with no date gives a board nothing to build its agent controls toward.
Aalok Mehta of the Center for Strategic and International Studies set two conditions for cooperation, and the first one caps what any text can contain. "The first is that it can't hurt the ability for the U.S. and China to compete on the merits of their technology," he told CNBC [16]. A clause touching training runs or launch dates fails that test before anyone drafts it.
The second condition is harder. "Policymakers won't be inclined to take their counterparts just on their word, so we'll need to find ways to ensure information flowing between the two countries is accurate," Mehta said [17].
Count what is on the table: a shared vocabulary for powerful models and an emergency line. Neither of those is binding when a model ships [22].
The chip demands are the clearest ones on the table. Jamieson Greer, the U.S. trade representative, said controls on chips were not on the agenda during weekend preliminary talks [13]. Melanie Hart of the Atlantic Council's Global China Hub told reporters on Tuesday that "China's asking the U.S. to embrace Chinese AI models and drop some of the U.S. controls on China's access to semiconductor chips" [14]. Several Chinese firms have reportedly reached the restricted chips' compute power remotely, through data centers in Southeast Asia [12]. If the compute is rentable across a border, a licence concession is worth less than the licence list implies, and a safety channel costs Washington less to hand over.
The talks produce definitions and nothing else. Or Beijing works the new channel as an export-control grievance, which is what Chris McGuire of the Council on Foreign Relations said China has historically used AI dialogues for, "to complain about U.S. export controls" [9]. Or an agent incident large enough to need the line arrives before any verification exists. I would weight the second highest, because McGuire also said Chinese state media "has largely rejected the concerns expressed by leading U.S. AI labs about the potential for catastrophic risk, which is consistent with China's long-held skepticism of all U.S. arms control proposals" [10].
The case against that weighting is commercial. If Beijing wants common safety definitions its own vendors can certify against, the definitional annex lands early, and companies running agents get something to log against while the chip fight continues. Chinese models have made big capability gains in recent months and are increasingly being adopted by companies globally, including in America [21].
What to watch
- Whether Thursday's meeting produces any joint text defining a powerful model, and which side drafts the definition.
- Whether chip export licences surface in a summit deliverable after Greer said they were off the preliminary agenda.
- Whether Bessent's July distillation sanctions threat becomes an actual designation against a named Chinese lab.