US and China talk about AI safety while trading theft accusations

In short: The US and China are trying to reduce real world harms from today’s AI systems, but the relationship is strained by mutual accusations and threats.

What's going on

US and Chinese officials are not only debating what AI could do in the future. Recent reporting says they are also trying to deal with problems that are already happening, like AI model theft claims, misuse by nonstate actors (groups that are not governments), and safety risks from very powerful “frontier” AI systems (the biggest and most capable models).

On the cooperation side, US officials have said the two countries are discussing “AI guardrails,” meaning basic safety rules and practices. One goal is to make it harder for nonstate actors to exploit advanced AI models. Think of it like agreeing on seatbelts and speed limits before letting more people drive faster cars.

China has also said it is willing to hold AI talks with the US. At the same time, Chinese officials have urged Washington to stop what they call “unfounded accusations,” and they have warned of retaliation if the US moves against Chinese companies over alleged technology theft.

Meanwhile, some US officials have accused Chinese firms of “industrial-scale” distillation, which is a way to copy what a top AI model can do by training another model on its outputs (like learning a chef’s signature dish by tasting it many times). China has pushed back strongly.

What to watch

Policy analysts say practical steps may matter more than big promises, like shared safety guidelines, limited information sharing about misuse, and emergency communication channels. The key question is whether the two governments can keep those narrow safety talks going while tensions rise over accusations, sanctions, and export controls.

Source: NYTimes