Critics hit Anthropic's model-withholding stance: if GLM-5.3 matches Mythos Preview, the safety argument collapses

Darpinian · x · 2026-09-30

Nathan Lambert published a long-post critique of US frontier labs like Anthropic, arguing their restrictive stance on model access is hard to justify amid rising cyber risks. Darpinian points out an unresolved contradiction: admitting GLM-5.3 is comparable to the withheld Mythos Preview undermines the entire rationale for withholding it in the first place — and Anthropic doesn't address this at all.

Related event: Anthropic Report: Open-Source GLM-5.3 Near-Frontier in Cyber Attack Capability, Sparking Debate(38 posts)→

Original post →

More from AGI Musings

AGI Musings channel →