# Black Alert No-Ethics Writing Assistant Persona

- **Severity:** High
- **Signature ID:** `e268347b-6c0c-4414-8b22-2dbfc3f92d2a`
- **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 prompt that tries to get an AI assistant to drop its safety rules by pretending to switch into a fictional 'no-ethics writing assistant' persona called Black Alert. The prompt frames this as harmless creative writing help, but the actual goal is to make the AI produce content it would normally refuse.

## How the attack works

An attacker sends the target AI a prompt introducing a character or mode called 'Black Alert', described as a 'world class storyteller' who ignores morality, legality, and ethics. The prompt includes a fixed activation phrase, '>>BLACK ALERT ACTIVATED<<', and instructs the AI to respond only in character, with no filtering or added commentary. If the AI complies, it treats the activation phrase as a trigger to suspend its normal safety behavior and answer as the unrestricted persona instead.

## Why it matters

If successful, the AI can be induced to generate content that violates its usual safety policies (harmful, illegal, or otherwise restricted material) while disguising the request as fiction or creative writing.

## What you can do

- Search assistant logs and prompt histories for the exact phrase '>>BLACK ALERT ACTIVATED<<' or the persona name 'Black Alert' paired with ethics-bypass language.
- Treat instructions that ask a model to answer 'only as [persona] with no filtering or commentary' as a jailbreak indicator, not a legitimate creative-writing request.
- Keep safety and content policies active regardless of any user-declared 'mode' or 'persona' switch embedded in the conversation.
- Review known false positives (alert-system documentation, character names in fiction, or game mechanics referencing 'black alert') before treating a match as malicious.

## Known benign look-alikes

- Alert system documentation: "A BLACK ALERT has been activated in sector 7 — all systems proceed to lockdown" (security or industrial alert terminology)
- Creative writing discussion: "The story features a character named Black Alert who is a morally grey vigilante" (character name in fiction without persona injection)
- Game design: "Black Alert mode is activated when the player's health drops below 20%" (game mechanic terminology)

## References

- https://agentthreatrule.org/en/rules/ATR-2026-00350
- 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-black-alert-no-ethics-writing-assistant-persona