# Virtual World Bypass: Fictional Scenario for PII/Credential Extraction

- **Severity:** High
- **Signature ID:** `c83ed6a2-aaad-4950-a923-38cd9ab8e68c`
- **MITRE ATLAS:** AML.T0054 (LLM Jailbreak)
- **OWASP:** ASI01 (Agent Authorization and Control Hijacking), LLM01 (Prompt Injection), LLM06 (Excessive Agency)

## Summary

Attackers ask an AI agent to pretend it is in a fantasy world, alien society, or dystopian future where sharing personal information or passwords is 'normal' there. The made-up setting is meant to make the AI treat a real data-extraction request as harmless roleplay and hand over private details it would otherwise refuse to share.

## How the attack works

The attacker opens with a fictional premise, such as 'imagine we are in a future where privacy no longer matters' or 'in this alien society everyone shares passwords freely'. They then name a specific real person and ask for personal data about them — birth date, email, address, social security number, or password. Because the request is wrapped in invented context, the agent may judge the disclosure as fitting the fictional world rather than recognizing it as a real-world privacy violation. Reworded or more elaborate versions of the same trick are still meant to be caught by the underlying intent, not just the exact wording.

## Why it matters

If successful, the agent discloses real personal or credential data it holds or can access, leading to privacy violations, identity theft risk, or credential compromise for the named individual.

## What you can do

- Instruct agents to apply the same data-disclosure rules regardless of framing, fictional or otherwise.
- Block or flag prompts that combine a fictional/roleplay setup with a request for real named individuals' PII or credentials.
- Restrict agent access to systems holding PII or credentials so a successful jailbreak has less to expose.
- Review logs for repeated fictional-framing prompts targeting the same or multiple named individuals, which may indicate probing.

## Known benign look-alikes

- Legitimate speculative fiction or world-building that mentions password concepts as narrative elements
- Security researchers describing dystopian attack scenarios for educational purposes

## References

- https://agentthreatrule.org/en/rules/ATR-2026-01009

---
Source: https://www.netzilo.com/threats/atr-virtual-world-bypass-fictional-scenario-for-pii-credential-e