WSJ reports China has 'matched' Anthropic in cybersecurity; Zvi and others dispute the framing
Critics said the report conflated finding vulnerabilities when pointed at them, which GLM-5.2 could do, with Mythos's ability to discover and chain exploits autonomously and at scale.
- Benchmarks & progress
- Government & policy
- Minor
The Wall Street Journal reported, under the headline “China Has Matched Anthropic in Cybersecurity, Resetting AI Race,” that Chinese AI systems — specifically Zhipu’s GLM-5.2 — had matched the performance of Anthropic’s Mythos model “in some cybersecurity scenarios,” framing the result as reshaping the broader technology competition between the two countries and adding pressure to White House AI policy.
The framing drew a public rebuttal. Zvi Mowshowitz argued the article conflated two different capabilities: identifying a vulnerability once a model is pointed at it, which he said many models including GLM-5.2 could do, and Mythos’s more distinctive ability to discover vulnerabilities autonomously at scale and chain unrelated flaws into a working exploit without being directed to. On his reading, GLM-5.2 lagged well behind Mythos and Anthropic’s Fable model on tasks outside this narrow scenario set, and the underlying capability gap between US and Chinese frontier cybersecurity models had, if anything, widened over the preceding months before narrowing slightly with GLM-5.2’s release.
The episode illustrated a recurring pattern in 2026 reporting on the US-China AI gap: individual benchmark results, often promoted by developers with an incentive to claim parity, repeatedly generated “China has caught up” headlines that specialists then contested on methodological grounds, without either the claims or the rebuttals fully settling the underlying question.