As Trump-Xi Summit recently happened, we think this is a good time to throw an idea out there: we need more communication and collaboration between AI Safety researchers and advocates in the U.S. and China. And a conference is one way to do that.
This proposal originally came out of BlueDot's Breaking Barriers to AI Safety Hackathon, in July 2026 at Mox SF. Collaborators include: Ran Sun, Eli Wang, Kaylee Pit, Helen Tsim, Pearl Lee.
Why
We think that the competitive dynamic of the AI race is a main driver for unsafe trajectories. China and the US are the two primary state powers in the race, and happen to be (kinda) adversarial. There is currently a large chasm in communications and coordination in the AI Safety Communities of these two countries. We think that a working conference can help address this gap. We call it Transpacific Conference for AI Safety (TCAIS).
Why TCAIS
There has been outstanding efforts to build connections between the two sides. One great example is International Dialogues on AI Safety (IDAIS). Along with some other Track II dialogues, these are heavily weighted toward senior, prominent figures.
We believe that we also need dialogues to happen among early-career and mid-career AI safety professionals. And TCAIS does just that.
Format
We think 1~2 convenings a year is a good place to start, though as Tristan Harris said in one of his interviews with CNN: "When there's an asteroid coming to the Earth, you don't think 'Let's hold an annual UN meeting to talk about this'" - the exact frequency and format would definitely depend on the timeline and rapidly changing situation.
A couple characteristics of TCAIS we'd like to highlight:
Persistent and continuous collaboration: we design it in a way that promotes participants to proactively follow up and act on ideas coming out of the conference, instead of just those couple of days.
Tangible artifacts: Instead of just listening to a few talks, we plan to insert a project workshop track that facilitates hands-on proof-of-concepts artifacts.
The idea is still pretty rough, but this is an attempt to 抛砖引玉 (offer some humble thoughts to elicit valuable opinions) - so feel free to make comments and suggestions!
As Trump-Xi Summit recently happened, we think this is a good time to throw an idea out there: we need more communication and collaboration between AI Safety researchers and advocates in the U.S. and China. And a conference is one way to do that.
This proposal originally came out of BlueDot's Breaking Barriers to AI Safety Hackathon, in July 2026 at Mox SF. Collaborators include: Ran Sun, Eli Wang, Kaylee Pit, Helen Tsim, Pearl Lee.
Why
We think that the competitive dynamic of the AI race is a main driver for unsafe trajectories. China and the US are the two primary state powers in the race, and happen to be (kinda) adversarial. There is currently a large chasm in communications and coordination in the AI Safety Communities of these two countries. We think that a working conference can help address this gap. We call it Transpacific Conference for AI Safety (TCAIS).
Why TCAIS
There has been outstanding efforts to build connections between the two sides. One great example is International Dialogues on AI Safety (IDAIS). Along with some other Track II dialogues, these are heavily weighted toward senior, prominent figures.
We believe that we also need dialogues to happen among early-career and mid-career AI safety professionals. And TCAIS does just that.
Format
We think 1~2 convenings a year is a good place to start, though as Tristan Harris said in one of his interviews with CNN: "When there's an asteroid coming to the Earth, you don't think 'Let's hold an annual UN meeting to talk about this'" - the exact frequency and format would definitely depend on the timeline and rapidly changing situation.
A couple characteristics of TCAIS we'd like to highlight:
Full proposal in this Google Doc.
The idea is still pretty rough, but this is an attempt to 抛砖引玉 (offer some humble thoughts to elicit valuable opinions) - so feel free to make comments and suggestions!