LLMs are better at attacking than writing secure code.

LLMs aren't terrible at securing systems, humans are bad at using secure systems and typically disable measures with insecure workarounds.