The Information Machine
Updated today·following since 3 Sep 2026·New·2 sources

China's shikong AI risk framework

The gist

China's 'Shikong' Organizes AI Risk by Actor, Not Harm Severity

Shared vocabulary between China and the West on AI safety does not mean shared conceptual frameworks; the same term structures risk around different actors. Observers who assume rhetorical overlap signals genuine alignment risk misreading where agreement actually exists.

The full picture

Analysis by Zilan Qian and the Oxford China Policy Lab finds that China's concept of 'shikong' (loss of control) covers three distinct scenarios: an operator losing control of a system, the state losing control, and permanent human loss of control over AI. The organizing principle in Chinese discourse is who is presumed to exercise control and who loses it, rather than the severity of harm or the mechanism of failure. The analysis also finds that China lacks discourse about loss-of-control risks within internal lab deployments, an area the analysis identifies as one of the biggest dangers.

How it developed
5 September 2026

Zilan Qian and the Oxford China Policy Lab published an analysis on September 4 finding that Chinese uses of 'shikong,' or AI loss of control, are organized by who loses control rather than harm severity, covering three loci: operator, state, and humanity.

Qian warns against assuming rhetorical overlap with Western AI safety terms signals genuine alignment, and the analysis identifies a specific gap: China lacks discourse about loss-of-control risks within internal lab deployments, which Qian describes as one of the biggest dangers.

4 September 2026

Transformer News covers Qian's warning against assuming China-West rhetorical alignment on AI safety

3 September 2026

Oxford China Policy Lab and Zilan Qian analysis of 'shikong' published

Sources
The daily email

Want this in your inbox?

I send a short email each morning with the stories that moved. If you would rather just read here, that works too.

Subscribe free