US and China start limited talks on shared AI safety risks

In short: The United States and China have started small, early talks about AI safety, even as their broader rivalry keeps growing.

What's going on

The basic idea is that Washington and Beijing can disagree on trade and technology, but still work together on a short list of AI dangers that could hurt both countries. One immediate concern is that powerful AI could help criminal or terrorist groups carry out cyberattacks, scams, or other harmful actions. A longer-term worry is losing human control over very capable AI systems, meaning the systems might behave in unexpected and dangerous ways.

There are a few early signs of contact. After a Trump and Xi summit, the two governments agreed to begin an AI dialogue. Chinese officials say the first round happened in September, another is planned for November, and both sides agreed to set up a channel to communicate about AI incidents. You can think of this like setting up a hotline, so there is a clear way to report and discuss problems.

Analysts described the progress so far as modest. They also warned that distrust and competition could limit how far cooperation goes.

What to watch

Researchers at Brookings argue that a practical next step is regular meetings between technical experts. These talks could focus on how to test AI models, “red-teaming” (trying to break a system like a friendly stress test), cybersecurity misuse, weapons-related risks, and major reliability failures. The same experts caution that these discussions should stay narrowly focused on shared technical risks, instead of turning into a broader fight about export controls, open-source rules, or the overall US-China rivalry.

Source: NYTimes