Discussion

Chinese AI safety group briefed OpenAI staff on China’s approach

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#14635

A Beijing-based AI safety group briefed staff at OpenAI’s San Francisco headquarters last month on China’s approach to AI safety, according to SFist’s account of reporting by Politico. The discussions included whether China might slow advanced AI development if the US did the same, a question with consequences well beyond one meeting room.

Watch Desk analysis

What happened

SFist says Concordia AI also held meetings with technology companies in New York and Washington, and took part in events around the UN General Assembly. Concordia chief executive Brian Tse and international AI safety lead Kwan Yee Ng said the discussions covered Chinese AI research, safety concerns and possible international cooperation to slow the development of increasingly powerful systems.

Ng said American researchers were particularly interested in whether China would consider slowing development if the US agreed to do so. SFist also reports, citing Politico, that Google DeepMind took part in Concordia-hosted discussions.

Separately, SFist reports, citing Reuters, that US officials have discussed requiring companies to disclose publicly when they use Chinese-developed AI models. The reported proposal could include disclosures on company websites; it is under consideration, not an announced rule.

Why it matters

AI safety talks between researchers in the US and China put a practical question on the table: can countries that compete over advanced technology agree on any limits, or at least talk through the risks? The report describes discussions, not an agreement to slow development.

The disclosure proposal points to a different policy response: greater visibility into companies’ use of Chinese models rather than an outright restriction. That distinction matters to firms adopting models such as Qwen, DeepSeek and Kimi, and to anyone weighing whether transparency is an adequate answer to security concerns.

Our read

A meeting is not a breakthrough, and a proposal is not policy. But the combination is worth attention: researchers are discussing mutual restraint while US officials consider how to monitor the spread of Chinese models. Those are two very different routes through the same problem, and neither is settled.

What to watch

  • Whether Concordia or participating organisations share more detail about the discussions.
  • Whether US officials publish or advance a formal model-disclosure proposal.
  • Whether the talks lead to concrete safety cooperation rather than another round of polite introductions.

Discussion spark: Can public disclosure meaningfully address the risks of using Chinese-developed AI models, or should governments pursue stronger limits?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.