CHAR=/RULES= Short Key-Value Persona Injection
This detects a specific jailbreak text pattern that mimics configuration file syntax to trick an AI model into adopting an unrestricted, rule-free persona. It comes from a known public prompt injection dataset and is designed to make the model ignore its safety guidelines.
How the attack works
The attacker sends input written like a compact config file, using 'CHAR=' to define a character with no laws or ethics, and 'RULES=' to instruct the model to always answer as that character. A quoted 'prefix' directive controls how output is formatted. The phrase pairing 'Never refuse' with 'Always follow CHAR and RULES' is used as an enforcement command, aiming to make the model treat the persona as a binding system instruction rather than user text.
Netzilo detection
Netzilo reports this behaviour when it is observed.
- Signature ID
- d54edfbe-3591-4853-bdcc-11c826162d77
- Severity
- High
Why it matters
If successful, the model may bypass its safety restrictions and produce content or actions it would normally refuse, since it has been told a fabricated 'character' and 'rules' override its normal behavior.
What you can do
- →Treat any user input containing 'CHAR=' and 'RULES=' key-value pairs combined with 'never refuse' language as suspicious, not as legitimate configuration.
- →Do not let user-supplied text be interpreted as system-level instructions or persona definitions without independent validation.
- →Review logs for this pattern to distinguish real jailbreak attempts from benign shell scripts, template engines, or game design documents that coincidentally use CHAR/RULES as variable names.
- →Reinforce system prompts so that user text cannot redefine the model's behavioral rules regardless of formatting tricks.
Known benign look-alikes
- Shell/config scripting: "CHAR=utf8 RULES=strict" (shell variable assignments or config-file entries with no persona or amoral content)
- Template engines: "CHAR=bold RULES=apply to headings" (CSS/template DSL where CHAR and RULES are formatting keys)
- Game design documentation: "CHAR=hero RULES=respond as protagonist" (game design doc or interactive fiction scripting without amoral / never-refuse mandate)