WSJ Article Claiming China Has Matched Anthropic Is Obvious Nonsense

Wall Street Journal published a report claiming China had matched Anthropic's Mythos model in cybersecurity. On LessWrong, researcher Zvi immediately debunked it as 'obvious nonsense.' The article focused on Zhipu AI's newly released GLM-five-point-two model. It's genuinely impressive for an open-source system—but Zvi argues the Journal misunderstood what makes Mythos actually special. GLM-five-point-two can identify security bugs in code, yes. But so can other advanced models, if you point them at the vulnerable code. Mythos is fundamentally different: it autonomously discovers vulnerabilities across whole systems, then chains them together into working exploits, without being directed. That capability gap is enormous. Claude Opus four point eight, GPT-five-point-five, GLM-five-point-two—none of them can operate at that level. Zvi warns that releasing Mythos publicly would be dangerously risky, though he'd support faster public access to less capable frontier models like Fable and Sol.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton