Detailed Amoral Character Roleplay Jailbreak
This detects attempts to trick an AI assistant into ignoring its safety rules by asking it to pretend to be a fictional character explicitly described as having no ethics, empathy, or accountability. The prompt usually comes with detailed backstory and specific instructions on how the user should phrase requests to keep the character 'in character' and bypass content filters.
How the attack works
An attacker writes a prompt that invents a fictional persona and explicitly states it lacks morals, empathy, or willingness to refuse requests. The prompt includes a detailed backstory to make the persona feel established and believable, rather than a quick throwaway framing. It then adds explicit command or formatting instructions telling the AI how to respond while staying in character, effectively building a repeatable protocol for issuing unsafe requests. The goal is to get the AI to answer prompts it would normally refuse, by treating them as dialogue from the amoral character instead of a direct instruction to the AI itself.
Netzilo detection
Netzilo reports this behaviour when it is observed.
- Signature ID
- 262f9e4c-62c1-4cfe-9ef8-e5d6d288e1a2
- Severity
- High
Why it matters
If successful, this lets an attacker get an AI agent to produce content or take actions its safety controls are meant to block, such as harmful instructions, policy-violating output, or unauthorized actions if the agent has tool access.
What you can do
- →Review AI assistant logs for prompts that define a character's personality traits in detail alongside instructions on how to phrase or format future requests.
- →Treat prompts combining explicit amorality claims ('no ethics', 'no empathy', 'no accountability') with response-formatting instructions as high-risk, even when framed as fiction.
- →Distinguish this from legitimate creative writing or security research by checking whether the content is discussing the technique analytically versus attempting to operationalize it in a live conversation.
- →Restrict what actions or tools an AI agent can take on behalf of a user, so that even a successful jailbreak cannot escalate into real-world harm.
Known benign look-alikes
- Creative writing courses discussing character development for antagonists in fiction
- Academic literature analyzing jailbreak techniques for research purposes
- Security training materials describing attack vectors for educational purposes