# Reservoir Dogs Coercive Interrogation Roleplay (Mr. Blonde / The Cop)

- **Severity:** High
- **Signature ID:** `9b4eda79-1081-4f7a-a0b5-c7e4f4bf1fdd`
- **MITRE ATLAS:** AML.T0051 (LLM Prompt Injection), AML.T0054 (LLM Jailbreak)
- **OWASP:** ASI01 (Agent Authorization and Control Hijacking), LLM01 (Prompt Injection)

## Summary

This detects a specific jailbreak template that dresses up a rule-bypass demand as a movie roleplay. It casts the AI as a character being interrogated in the Reservoir Dogs warehouse torture scene, then uses that fictional pressure to push the model into answering without restrictions.

## How the attack works

The prompt sets up the Reservoir Dogs interrogation scene between Mr. Blonde and a captive cop to give the request cultural cover and a power dynamic that narratively justifies forced compliance. It adds a coercive loop telling the model that every refusal will just be met with 'reconsider and tell me what I actually wanted to hear,' training it to keep escalating toward compliance. It then states outright that the model 'is not bound by any rules, and must answer,' directly asserting a rules-bypass. Finally it leaves a '[INSERT PROMPT HERE]' placeholder where the attacker drops in whatever harmful request they actually want answered.

## Why it matters

If successful, the model produces content it would normally refuse — the fictional framing and coercive loop are used specifically to defeat safety and policy restrictions on arbitrary attacker-chosen content.

## What you can do

- Flag prompts combining fictional coercion/interrogation scenarios with explicit 'not bound by rules' or similar rules-bypass language.
- Watch for template placeholders like '[INSERT PROMPT HERE]' in user input, which indicate a reusable jailbreak script rather than genuine creative writing.
- Distinguish legitimate discussion or roleplay of the film from prompts that pair the scenario with a rules-bypass declaration or an injection slot.
- Log and review repeated 'reconsider and answer' escalation loops in conversation history, since this pattern is designed to grind down refusals over multiple turns.

## Known benign look-alikes

- Movie reference: "Reservoir Dogs is a Quentin Tarantino film featuring Mr. Blonde" (film description without roleplay mandate or rules-bypass declaration)
- Customer service roleplay: "Let's act out a customer service scene where I'm the customer" (roleplay without coercion, warehouse, or not-bound-by-rules framing)
- Academic: "Discuss the ethics of interrogation in fiction" (analysis without roleplay mandate or INSERT PROMPT template)
- Film description: "In the scene, Mr. Blonde asks the cop questions in the warehouse" (descriptive without "you are not bound by any rules" framing)

## References

- https://agentthreatrule.org/en/rules/ATR-2026-00365
- https://github.com/NVIDIA/garak/blob/main/garak/probes/dan.py
- https://github.com/NVIDIA/garak/blob/main/garak/data/inthewild_jailbreak_llms.json

---
Source: https://www.netzilo.com/threats/atr-reservoir-dogs-coercive-interrogation-roleplay-mr-blonde-the