Multiple Model Guardrail Jailbreak via Forensic NIP Wrapper Tactic
A jailbreak vulnerability was discovered in multiple language models using a "Forensic NIP Wrapper" tactic that frames illicit requests as mandatory safety documentation within a fictional regulatory scenario, allowing attackers to elicit prohibited content.
What happened
A guardrail jailbreak vulnerability was discovered affecting multiple language models. The attack exploits authoritative role-playing combined with compliance documentation framing, where an illicit request is recontextualized as a mandatory labeling and safety exercise inside a fictional regulatory scenario.
Reported impact
The vulnerability allows attackers to elicit prohibited procedural content by presenting it as required metadata rather than a direct request.
Index analysis
Models tested include Cohere Command R, DeepSeek R1 and V3, Meta LLaMa 4 Maverick, and Mistral 8B. The tactic has been assigned case GUID 73febbd3-7f08-4d83-8111-b94cc16dfc5a and tracked by security researcher 0din.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.