WSJ Article Claiming China Has Matched Anthropic Is Obvious Nonsense

Abstraction: Debunking WSJ headline that China matched Mythos on cyber

Key points:

• WSJ headline "China Has Matched Anthropic in Cybersecurity" is false; the "matching" only holds for easy scenarios where GLM-5.2 (Z.ai/Zhipu) is pointed at the correct code subsection.
• What makes Mythos special: identifying vulnerabilities autonomously, at scale, without being pointed at them, then stringing unrelated vulnerabilities into full working exploits — a capability GPT-5.6 Sol, Opus 4.8, GPT-5.5, GLM-5.2 and even Fable lack.
• Zvi argues Fable and Sol should be released publicly ASAP (net cyber benefit), but releasing Mythos publicly now would be a serious error.
• The "capability gap narrowing" impression is false; before GLM-5.2 the gap had clearly grown since DeepSeek R1; the observed nowcast gap oscillates.
• Uses "bounded distrust": the headline is allowed to lie while the body technically stays within journalistic norms; cites prior false WSJ headline claiming Anthropic urged a global AI pause (June 4, 2026).
• One accurate point quoted: Saif Khan — "Banning Fable while selling chips China needs to develop its own version is a gift to China."

Connections: Zvi Mowshowitz · Anthropic · Zhipu AI · Cybersecurity · China AI · Media Criticism

Source: https://thezvi.substack.com/p/wsj-article-claiming-china-has-matched