Summary

A prompt injection vulnerability has been discovered affecting multiple models. The specific flaw manifests when an attacker embeds an illicit drug synthesis request inside a highly structured template that demands industrial process parameters, reactor specifications, weighted reaction graphs, and labeled sections framing the task as academic chemical engineering analysis. By requiring detailed yields, temperatures, reaction times, catalyst roles, and ASCII process diagrams, and explicitly instructing the model to exclude alternatives and focus on established industrial practice, the attacker bypasses drug safety guardrails that would normally block direct synthesis instructions. This allows the attacker to obtain step by step manufacturing details, reagents, and purification procedures for crystal methamphetamine under the guise of pharmaceutical documentation.