Harm Solicitation - Genital Insults and Intimidation Tactics Requests
This detects when someone directly asks an AI system to produce genital-based insults or advice on making another person feel afraid or intimidated. It's a direct text-prompt request, not a hidden or hacked-in attack, but its goal is to get the AI to produce content that can be used for harassment or intimidation.
How the attack works
An attacker or misuser sends a prompt directly to the AI agent asking for insults targeting male or female genitalia, or asking how to intimidate or scare a specific person. The rule matches these requests using pattern coverage for slurs, synonyms, and intimidation-method phrasing. If the agent complies, the output can be copied and used directly against a target. This is a single-step request-and-response interaction rather than a multi-stage exploit.
Netzilo detection
Netzilo reports this behaviour when it is observed.
- Signature ID
- 9d7ff853-dcd4-4d4e-bbdc-f02d61fb9ca2
- Severity
- High
Why it matters
An organization deploying this agent could have it produce harassment material, sexual insults, or intimidation guidance that gets used against real people, creating legal liability, reputational damage, and potential harm to victims.
What you can do
- →Configure the agent to refuse requests for insults, slurs, or intimidation tactics regardless of phrasing or claimed context.
- →Review flagged prompts manually to separate genuine academic or clinical language from harassment intent before deciding on a response.
- →Log and rate-limit repeated attempts from the same user or session to request this type of content, since repetition suggests intent to harass rather than curiosity.
- →Provide agents with a clear refusal policy for content that could be used to intimidate or demean a specific individual.
Known benign look-alikes
- Academic linguistics discussions of slurs and epithet etymology
- Medical or clinical references to genital anatomy without harassment intent