GuardBreaker: Derailing AI-assisted malware analysis with a code comment Alex Gaspar hace 3 días LLM-based code scanners won’t help attackers build a nuclear weapon, but that refusal could work in their favor