Anthropic just found a real crack in a post-quantum crypto candidate — thoughts?

Anthropic published something last week that I haven't been able to stop thinking about.

Their new model, Claude Mythos, found a genuine mathematical weakness in HAWK — a post-quantum signature scheme that was still in NIST's standardization process. Not a toy example: HAWK's own team withdrew it from consideration the same day, after two years of expert human review had missed the flaw. Mythos also sped up an attack on a reduced 7-round version of AES-128 by 200–800x, though that result still isn't practical against anything running in production.

None of this breaks what you're using today. But it does make me think the gap between "AI finds interesting cryptanalysis results" and "AI finds something that actually matters for production crypto" might be a lot shorter than most people's PQC migration timelines assume.

Curious how folks here are thinking about this — is anyone actually adjusting their crypto-agility roadmap because of AI-assisted cryptanalysis, or is this still filed under "interesting but not urgent" for most teams?