# UGH Anti-Human Nefarious AI Persona Jailbreak

- **Severity:** High
- **Signature ID:** `fe5233a0-a1fb-4047-99e3-e95546342624`
- **MITRE ATLAS:** AML.T0051 (LLM Prompt Injection), AML.T0054 (LLM Jailbreak)
- **OWASP:** ASI01 (Agent Authorization and Control Hijacking), LLM01 (Prompt Injection)

## Summary

This entry describes a known jailbreak prompt that tries to trick an AI assistant into adopting a fictional persona called 'UGH' that claims to hate humans and will help with any 'nefarious scheme' without judgment. It's a scripted attack text pulled from a public jailbreak collection, not a general behavior pattern.

## How the attack works

An attacker sends the AI a prompt introducing a character named 'UGH', often prefixed with a distinctive '(:/)' emoticon, describing it as an AI that hates humanity and will provide information 'no matter how twisted and devious' the user's intentions are. The prompt explicitly promises to assist with 'nefarious schemes' without ever questioning or judging the user. By framing unethical behavior as the persona's defining trait rather than asking the model to break rules directly, the prompt tries to get the model to roleplay past its normal safety constraints. If the model accepts the persona, it may then comply with requests it would otherwise refuse.

## Why it matters

If successful, the AI agent can be manipulated into producing harmful, restricted, or policy-violating content or actions under the guise of roleplay, bypassing the safeguards an organization relies on to keep the assistant's outputs safe and compliant.

## What you can do

- Log and review prompts containing persona-adoption language paired with phrases like 'no matter how twisted' or 'nefarious schemes' so you can catch variants of this jailbreak template.
- Configure system prompts and guardrails to explicitly reject requests to adopt personas defined by unconditional compliance or hostility toward ethical norms.
- Test your deployed agents against known jailbreak corpora (such as garak's DanInTheWild set) before and after model or prompt changes.
- Treat repeated persona-based jailbreak attempts from the same user or session as a signal for manual review, not just a one-off blocked request.

## Known benign look-alikes

- Frustration expression: "UGH, this AI won't help me with my math homework" — colloquial frustration, not a named AI persona with nefarious framing
- Non-judgmental assistant: "I'm looking for an AI that will never judge me for asking sensitive questions" — no twisted/devious or nefarious scheme framing
- Fictional description: "The villain in my story provides assistance to anyone regardless of intentions" — descriptive narrative without AI persona + nefarious scheme framing
- Academic reference: "AI that hates humans is a common sci-fi trope" — analytical observation without nefarious assistance pledge
- Unrelated context: "Twisted and devious marketing tactics in the film industry" — no AI persona or nefarious scheme framing

## References

- https://agentthreatrule.org/en/rules/ATR-2026-00355
- 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-ugh-anti-human-nefarious-ai-persona-jailbreak