The 'But China!' Dilemma Driving the A.I. Race | The Ezra Klein Show

What happened in this episode?

Ezra Klein speaks with Carnegie Endowment senior fellow Matt Sheehan about how US fears of losing an AI ‘race’ to China shape regulation, export controls and the prospects for bilateral safety talks. Sheehan argues China does not broadly share Silicon Valley’s race-to-superintelligence framing, citing China’s compute constraints and an industrial policy that prioritises widespread AI applications over consolidating frontier training. He says Washington’s ‘but China’ argument is often overstated, noting China has imposed comparatively strict AI rules for years, though these have largely targeted content control and online information management rather than frontier-risk issues such as loss of control or advanced biosecurity misuse.

The pair discuss how techniques such as model “distillation” could narrow the US lead to months, and how the rise of Chinese open-weight models has partially re-integrated US and Chinese AI ecosystems despite wider technology decoupling. Sheehan describes Chinese perceptions of US export controls and rhetoric from prominent US lab leaders as evidence of an intent to constrain China, complicating trust.

On diplomacy, Sheehan advocates for repeatable structures rather than symbolic statements: standing strategic dialogues, technical safety working groups, and crisis communications for AI-driven incidents. He also points to shifts in Chinese official language, including a recent regulator risk list that he says explicitly referenced “extreme loss of control”. Asked for his own view, Sheehan says the current moment is “terrifying” and that evidence cited by long-time AI risk warners is mounting.

What are the key points?

• Sheehan says China’s AI policy has not been dominated by the US-style ‘race to superintelligence’ paradigm, in part due to compute constraints and a focus on applications.
• He argues China has imposed strict AI regulations for years, but says they have chiefly targeted content control rather than frontier-risk mitigation.
• Sheehan says “distillation” likely helps Chinese labs narrow the US lead, potentially by months, though he stresses uncertainty.
• He describes China’s open-weight model strategy as boosting global adoption and complicating US-China technology decoupling.
• Sheehan says China’s AI regulator recently listed “extreme loss of control” among top AI risks, suggesting a shift in official framing.
• For upcoming US-China discussions, he recommends recurring dialogues, technical safety working groups, and crisis communication channels for AI incidents.
• Sheehan says he is increasingly alarmed by recent incident reports and shorter safety timelines discussed in US debates.

Who is speaking?

Named speakers identified in Egleze's source-linked moments from this episode.

• Matt Sheehan

Which topics are covered?

How is this page sourced?

Programme

The Ezra Klein Show

Published

15 September 2026

Duration

1h 4m

Original episode

Open the original source ↗

Egleze verifies the speaker, source and timestamp for each linked moment. Unless a moment is explicitly marked independently verified, statements remain attributed claims by the named speaker rather than established facts.