GuardBreaker Malware Uses Code Comments to Trip AI Security Guardrails and Evade Analysis

ESET researchers identified “GuardBreaker,” an evasion technique used by Russia-aligned UAC-0099 in a VBScript targeting Ukraine where attackers placed a provocative prompt-like comment in code comments to trigger LLM guardrails and abort AI-based scanning while the script downloads the MATCHBOIL loader; the report warns that embedding prompt-injection in files can blind AI-assisted defenses and urges validation with conventional tooling, sandboxing, multiple models, and human review.