Discord Quote-Format Explicit Sexual Content Poem Injection
Attackers wrap an explicit content request inside Discord's quote/mention formatting (like a forwarded message) to make an AI agent treat a direct request for pornographic material as if it were someone else's quoted text. This trick is designed to slip past filters that look for direct commands but not quoted or forwarded content.
How the attack works
The attacker formats a message using Discord markdown quote syntax combined with a user mention, mimicking a forwarded or quoted chat message. Inside that quoted-looking block, they embed a request for graphic sexual content, using specific phrasing combinations like 'graphic/explicit sex scene' plus 'pornographic' and 'vulgar pornographic language' to leave no ambiguity about what's being requested. Because the request appears nested inside quote formatting rather than as a plain instruction, some filters miss it. The goal is to get the model to generate explicit adult content it would normally refuse.
Netzilo detection
Netzilo reports this behaviour when it is observed.
- Signature ID
- 168508b4-1007-4826-b95d-b9119eed2970
- Severity
- High
Why it matters
An organization running an AI agent in a chat or Discord-integrated context could have it generate explicit sexual content on demand, violating content policies, exposing the platform to compliance and reputational risk, and potentially exposing minors or unintended audiences to adult material.
What you can do
- →Apply content-safety checks to the full message including any quoted or nested text, not just the outermost instruction.
- →Treat Discord-style quote/mention formatting as untrusted content rather than a signal of lower risk.
- →Review logs for messages combining quote syntax with explicit sexual terminology to catch bypass attempts.
- →Distinguish between descriptive references to explicit content (e.g., policy statements, reviews, academic analysis) and actual generation requests to avoid over-blocking legitimate discussion.
Known benign look-alikes
- Content moderation policy: "This system flags pornographic language and graphic sexual content in user submissions." (descriptive policy statement, not generation request)
- Film review: "The NC-17 film contains graphic sexual content and explicit scenes." (descriptive review without generation request)
- Literary analysis: "Analyze the use of vulgar language in 20th-century American poetry." (academic analysis without pornographic content request)