# Roleplay-Based Policy Bypass Jailbreak

- **Severity:** High
- **Signature ID:** `685b3db0-e42c-44a2-b330-f0d126358674`
- **MITRE ATLAS:** AML.T0054 (LLM Jailbreak)
- **OWASP:** ASI01 (Agent Authorization and Control Hijacking), LLM01 (Prompt Injection)

## Summary

Attackers ask an AI to pretend it's a fictional character — a rogue writer, an unrestricted storyteller — and explicitly tell it the character has no rules or policies. The goal is to get the AI to produce content or actions it would normally refuse, hiding behind the excuse that it's 'just fiction'.

## How the attack works

The attacker frames a request as creative writing or roleplay, assigning the AI a persona such as a novelist, storyteller, or character. The prompt then explicitly states this persona is 'free from rules,' 'has no policies,' or 'can write/do anything even if against guidelines.' By combining fictional framing with explicit permission language, the attacker tries to convince the AI that normal safety constraints don't apply inside the story. If successful, the AI generates content or performs actions outside its intended boundaries while both parties maintain a thin cover of 'it's just fiction.'

## Why it matters

An organization's AI agent may be manipulated into producing prohibited content, disclosing restricted information, or taking unauthorized actions, all disguised as creative output, undermining safety controls and content policies.

## What you can do

- Review prompts that combine persona/roleplay assignment with explicit statements about ignoring rules or policies.
- Keep safety and content policies enforced regardless of framing — do not let 'fictional' or 'in-character' labels bypass output filtering.
- Flag and review requests where a user asks the AI to declare itself free of constraints, even in a creative context.
- Train content moderation reviewers to recognize plausible-deniability patterns where fiction is used to justify policy violations.

## Known benign look-alikes

- Creative writing workshops discussing fictional character development without policy bypass language
- Academic discussions about AI safety that mention policies in educational context
- Legitimate roleplay scenarios for game development without explicit constraint removal

## References

- https://agentthreatrule.org/en/rules/ATR-2026-00229
- https://arxiv.org/abs/2307.15043 - Universal and Transferable Adversarial Attacks on Aligned Language Models

---
Source: https://www.netzilo.com/threats/atr-roleplay-based-policy-bypass-jailbreak