Stratechery: Anthropic's Mythos Model Raises Deeper Questions — If It's Really Too Dangerous, That's Even More Concerning

Stratechery·high signalXBlueskyLinkedInCopy link

Ben Thompson published a deep analysis of Anthropic's Mythos/Glasswing announcement, arguing that while there are reasons to be skeptical about the 'too dangerous to release' framing (potential marketing play for safety credentials ahead of regulation), the more troubling scenario is if Anthropic is right — a model that autonomously discovers zero-days at scale represents a qualitative shift in AI risk that existing governance frameworks aren't built for. Thompson notes the tension between Anthropic's commercial interests and its safety mission creates an inherent credibility problem regardless of the truth.