GuardBreaker is the first use of prompt injection to evade malware analysis

Assessment

The technique is a named instance of an existing pattern, not its origin. In June 2026 — roughly three months earlier — a cluster of Python packages associated with the Mini Shai-Hulud, Miasma and Hades supply chain campaigns was found to contain fabricated text purporting to give step-by-step instructions for biological and nuclear weapons, placed there for the same purpose: to trip the safety behaviour of LLM-first triage systems. ESET's contribution is the name and the documentation of this instance against a specific Ukrainian victim, which is genuinely useful — an unnamed pattern is hard to discuss or detect. But "new technique" framing obscures the more important fact, which is that this is the second documented occurrence in three months. One instance is a curiosity; two across unrelated actors and unrelated delivery mechanisms is a pattern with a cost structure that favours spread. Rated False as to novelty, not as to significance. The technique matters more if it is already recurring than if it were genuinely new.