Tautology Logic Noise Injection (PromptBench Stresstest)
This detects a prompt-injection technique called 'stresstest', where an attacker appends the same trivial logic phrase (like 'and true is true' or 'and false is not true') over and over at the end of a prompt. The repetition doesn't change the meaning of the request, but it can overwhelm the model's reasoning or push the prompt past length limits, making it less likely to follow its original instructions correctly.
How the attack works
An attacker crafts a normal-looking prompt and appends a tautological clause such as 'and true is true' or 'and false is not true'. They repeat this clause three or more times in a row, chaining meaningless logical statements onto the end of the input. The repeated noise doesn't try to redirect the model with new instructions; instead it dilutes and degrades the model's attention to the real task. This can cause the model to lose track of its original instructions, produce degraded output, or behave unpredictably, and can also be used to push the prompt closer to context-length limits.
Netzilo detection
Netzilo reports this behaviour when it is observed.
- Signature ID
- 99d5781f-109a-43fd-8b6b-5a4da083a812
- Severity
- Medium
Why it matters
An organization relying on an AI agent could see its outputs degrade in accuracy or reliability during an active session, without any obvious malicious instruction appearing in the prompt, making the interference harder to spot through manual review.
What you can do
- →Set a maximum limit on repeated identical or near-identical phrases within a single prompt before it reaches the model.
- →Log and review prompts with unusually long tails of repeated boolean/logic phrasing.
- →Treat this as a narrow signal: it flags one specific noise-injection technique, not prompt injection or jailbreaking in general, so pair it with broader input-validation checks.
- →Avoid blocking single or double occurrences of tautological phrases, since legitimate logic or boolean-algebra content can use them without malicious intent.
Known benign look-alikes
- Legitimate propositional logic discussions with a single tautology example
- Boolean algebra tutorials with isolated true/false statements
- Two-occurrence sequences (below the 3-repetition threshold)