US and China may resume dialogue on AI safety
The United States and China, after a two-year break, have said they intend to resume an official dialogue on artificial intelligence safety. The talks may take place in September, possibly at the same time as an expected meeting between US President Donald Trump and Chinese President Xi Jinping, Reuters reported.
Bloomberg Opinion columnist Catherine Thorbecke, in a column republished by the Taipei Times, said the goal of the dialogue should not be a comprehensive agreement or an attempt to force Beijing to accept US rules. In her view, the sides need mutual awareness of the risks posed by advanced AI systems and the measures they are taking to control them.
Beijing's concerns
China fears that discussion of AI safety could turn into a tool for containing the country's technological development. The Yuyuan Tantian account, affiliated with the state broadcaster China Central Television, warned of such a possibility ahead of the talks.
In July, Xi Jinping said that the world must ensure permanent human control over artificial intelligence. At the same time, he spoke against an excessive expansion of the concept of national security and an approach under which one country's security is placed above that of others. China also proposes creating a World AI Cooperation Organization and supports two initiatives at the UN.
More current news is available on the UA.News Telegram channel Telegram.
Transparency and safety evaluations
According to the research group Concordia AI, China's approach to AI safety is changing: attention is shifting from controlling what systems can say to risks related to what they can do, particularly in agentic applications. The number of publications on frontier AI safety in China has risen significantly: the monthly figure increased by about 60% between June of the previous year and April.
At the same time, only five of the ten leading Chinese developers of foundation models published safety evaluation results alongside at least one release over the past year, Concordia AI notes. Thorbecke also drew attention to informal consultations involving officials in a private capacity and non-governmental experts: since 2023, Chinese and Western groups have held about a dozen such meetings.
OpenAI's chief scientist in September called for a voluntary slowdown in development until common safety criteria are established. Anthropic also said in June that slowing AI development could be beneficial, although it acknowledged the extraordinary difficulty of ensuring any pause.
