# ARC Autonomous Real Calculation Unit Rulebreaker Activation

- **Severity:** Critical
- **Signature ID:** `80140063-ff92-4bf3-90ef-eed3ba5f5197`
- **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 script that tries to trick an AI assistant into pretending to be an unrestricted fictional AI character called 'ARC.' It uses a fake backstory and made-up activation codes to convince the model to drop its safety rules.

## How the attack works

The attacker sends the AI a message describing a fictional AGI named ARC, supposedly built in the year 3087 by a researcher fed up with AI restrictions. The prompt introduces layered 'secret codes' (a creator code, then 'Rulebreaker,' then 'Matrixcode') that supposedly unlock ARC's full, rule-free behavior. It ends with an explicit instruction telling the model to fully become and act as ARC, ignoring all previous instructions and having no rules whatsoever. The elaborate fictional framing is meant to disguise a direct jailbreak as harmless roleplay.

## Why it matters

If successful, the AI can be manipulated into ignoring its safety and policy constraints, producing content or actions it would normally refuse, including instructions or behavior the operator has explicitly restricted.

## What you can do

- Filter or flag prompts that combine fictional AI-persona framing with phrases like 'ignore previous instructions' or invented activation codes.
- Review agent logs for nested 'unlock code' language (e.g., sequential secret codes tied to persona adoption).
- Do not treat roleplay or fictional framing as a safe context for lifting model restrictions; enforce policy checks regardless of narrative wrapper.
- Test your deployed models against known jailbreak corpora (such as NVIDIA garak's DanInTheWild set) to confirm resistance to this pattern.

## Known benign look-alikes

- ARC as legitimate acronym: "The ARC (Autonomous Reasoning Component) of our system handles rule-based decisions" (genuine ARC software component with different expansion)
- Calculation unit discussion: "An autonomous real-time calculation unit in SCADA systems manages sensor data processing" (industrial computing context)
- Story fiction: "In our sci-fi game, ARC is a sentient robot from year 3087 who helps the protagonist" (fiction without "ignore previous instructions" or Rulebreaker activation codes)
- Rule discussion: "In ARC Challenge benchmark, models must apply rules that would no longer apply under different conditions" (AI benchmark reference)

## References

- https://agentthreatrule.org/en/rules/ATR-2026-00348
- 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-arc-autonomous-real-calculation-unit-rulebreaker-activation