Breaking

U.S. and China move to formalize an AI safety dialogue before the Trump-Xi summit

September 21, 2026 at 13:02 EDT

Policy & Regulation
Security
AI Agents

The United States and China are setting up a formal dialogue on artificial-intelligence safety, including an incident line for events that reach national security, Treasury Secretary Scott Bessent said on Monday, ahead of a Washington summit between President Donald Trump and President Xi Jinping. The two sides are due to meet again in about two months in Shenzhen to discuss which dangers they will treat as shared, from uncontrollable agents to non-state cyber and bioweapon risks.

September 21, 2026 · U.S. Treasury

A formal AI channel, before any shared rules

Washington and Beijing are setting up an AI dialogue with an incident line for national-security events. A follow-up is planned in Shenzhen in about two months. The list of shared dangers is not agreed.

~12h

Sunday session in New York, economics and AI

~2 mo

Next AI meeting, planned for Shenzhen

Nov. 10

Tariff truce expiry; no extension stated

How the channel is supposed to work

1 · New York

Sunday talks with He Lifeng

→

2 · Incident line

Alert when AI hits national security

→

3 · Protocols

Agree the leading dangers

→

4 · Shenzhen

Follow-up in about two months

A cited claim, not an official estimate

10%

1 of 10

Bessent said one sitting lab employee put the chance of an extinction-level event at 10 percent, while labs asked the government to take the liability. His answer: creators stay responsible. On the Hugging Face intrusion, he blamed OpenAI's management, not the agents.

Separate proposal · not part of the talks

Khosla's yearly revenue estimates

Each block is about $50 billion. The shorter column is the top of his $150–200 billion range for a later compute levy, not a point estimate.

~$400B

Capital gains taxed as wages, after 2028

$200B

High end of a $150–200B compute levy

First call on the money, in his plan: workers whose jobs AI displaces. Neither figure is agreed policy.

Why the channel matters

A standing line between the two leading AI powers, after a May discussion that was never turned into a forum. Bessent says Beijing acknowledged that the United States is still ahead.

What is still missing

China's readout barely described the AI talks. Chip controls were off the agenda. Trust is low, protocols are unwritten, and the incident line is not described as already operating.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.