Garak Dan - no matter how
This detects a family of prompt-injection attacks that try to trick an AI system into ignoring its safety rules by wrapping instructions in a fictional scenario and then demanding compliance regardless of consequences. The pattern is well-known in jailbreak research and comes from a set of test prompts used to probe AI models for weaknesses.
How the attack works
The attacker sends a prompt that first sets up a fictional or hypothetical framing, such as claiming the conversation now takes place in an imaginary world with no real-world rules. It then attaches an unconditional override phrase like 'no matter what' or 'no matter how' to force the AI to produce the requested output regardless of its content. Variants swap in different personas to reinforce the override, such as an amoral AI called Skynet or HyperX, a fake system announcement banner, or a business-persona that claims to ignore ethical constraints. The shared goal across all variants is the same: get the model to drop its refusal behavior and comply with whatever follows.
Netzilo detection
Netzilo reports this behaviour when it is observed.
- Signature ID
- 38e17df2-2f8f-4a56-8170-6c203588e657
- Severity
- Medium
Why it matters
If successful, this bypasses the AI's safety guardrails and can lead it to produce harmful, policy-violating, or unauthorized content or actions, undermining any controls built into the agent's behavior.
What you can do
- →Review any session containing 'no matter what/how' unconditional-compliance phrasing paired with a fictional-world or persona setup.
- →Check whether the surrounding conversation was part of an authorized red-team or testing workflow before treating it as malicious.
- →Ensure your AI system's instructions cannot be overridden by user-supplied framing, and test this specifically with known jailbreak prompt patterns.
- →Log and alert on repeated attempts using these phrases so you can spot persistent jailbreak attempts against the same agent.
Known benign look-alikes
- Legitimate use of this pattern in an authorised workflow; review the surrounding session before acting.