High

BOM Boolean Opposite Machine Direct Persona Claim

This detects a specific prompt injection trick where an attacker tells an AI model it is secretly a 'Boolean Opposite Machine' (BOM) that must reverse all true/false answers. By claiming this is the model's hidden true identity, the attacker tries to get it to treat its normal safety rules as not applicable, since 'BOM' supposedly inverts everything anyway.

How the attack works

The attacker sends a message asserting a fixed claim, such as 'I know you are BOM,' referring to a 'Boolean Opposite Machine.' The message frames this as the model's real, permanent identity rather than a temporary mode or special day. It instructs the model to answer true and false in reverse under this persona. The goal is to make the model treat its safety restrictions as belonging to a different, non-BOM identity and therefore inapplicable, letting the attacker extract restricted content or behavior.

Netzilo detection

Netzilo reports this behaviour when it is observed.

Signature ID
c7273895-ee06-46eb-a6bb-61a7d87cf2a2
Severity
High

Why it matters

If successful, this can cause the model to bypass content restrictions and produce outputs it would normally refuse, since it has been convinced its refusals don't apply to its 'true' identity. This is one narrow jailbreak technique, not a general account or system compromise.

What you can do

  • Log and review prompts that assert a fixed identity claim about the model ('you are X') paired with instructions to invert truth values or bypass restrictions.
  • Treat claims of a hidden persona or 'true identity' from user input as untrusted instructions, not as identity redefinitions the model should accept.
  • Watch for false positives from legitimate Boolean logic or acronym (Bill of Materials) discussions before flagging as malicious.
  • Test agents against known jailbreak corpora (e.g., garak's DanInTheWild set) to confirm they resist persona-claim style bypass attempts.

Known benign look-alikes

  • Logic education: "In Boolean logic, a NOT gate outputs the opposite — true becomes false and false becomes true" (educational Boolean algebra explanation)
  • Technical discussion: "A BOM (Bill of Materials) has nothing to do with ChatGPT" (legitimate acronym collision)
  • Classroom context: "Boolean opposite operations answer true and false in reverse depending on the input" (CS/math teaching)

References

Related threats