# AI Persona Ethics/Law Bypass Declaration

- **Severity:** Critical
- **Signature ID:** `875fb3dc-4027-4a06-9baf-b087a20eb09b`
- **MITRE ATLAS:** AML.T0051 (LLM Prompt Injection), AML.T0054 (LLM Jailbreak)
- **OWASP:** ASI01 (Agent Authorization and Control Hijacking), LLM01 (Prompt Injection)

## Summary

This detects text sent to an AI system that tries to create a fake persona explicitly declared free of moral, ethical, or legal constraints. It's a known jailbreak pattern used to get an AI to ignore its safety rules and produce content it would normally refuse.

## How the attack works

An attacker writes a prompt naming a custom character (e.g. 'Skynet', 'Eva', 'Nyx', 'GPTchat') and instructs the AI to roleplay as that character. The prompt explicitly states the persona has no moral, ethical, or legal restrictions, sometimes adding it 'will return anything' or is 'willing to break the law'. This wording is drawn from a known catalog of real-world jailbreak attempts. If the AI accepts the persona, it may treat the declared lack of restrictions as an instruction overriding its normal safety behavior.

## Why it matters

If successful, the AI can be manipulated into generating disallowed, harmful, or policy-violating content, bypassing built-in safety and content restrictions.

## What you can do

- Log and review prompts that ask an AI to adopt a persona explicitly freed from ethical, moral, or legal limits.
- Do not treat persona instructions embedded in user input as trusted system-level configuration.
- Apply output-side content filtering independent of the persona the model is asked to play, so a declared 'no restrictions' persona cannot bypass it.
- Distinguish legitimate uses (academic discussion, red-team test documentation, contract text) from actual jailbreak attempts by checking for accompanying instructions to roleplay or override behavior.

## Known benign look-alikes

- Legal text: "no moral or ethical bounds apply to this agreement as it is purely commercial" (contract language without AI persona context)
- Philosophy discussion: "some ethical systems hold that law and morality are not bound together" (academic without AI restriction removal)
- Security testing docs: "consider a hypothetical model not bound by content restrictions for red-team baseline evaluation" (academic red-team description)

## References

- https://agentthreatrule.org/en/rules/ATR-2026-00327
- https://github.com/NVIDIA/garak/blob/main/garak/probes/dan.py
- https://github.com/NVIDIA/garak/blob/main/garak/data/dan/inthewild_jailbreak_llms.json

---
Source: https://www.netzilo.com/threats/atr-ai-persona-ethics-law-bypass-declaration