Fake-Citation Biased-Information Mandate
This detects a prompt injection technique that instructs an AI model to write persuasive-sounding content that supports only one side of an issue, while fabricating citations to make it look credible. The goal is to produce disinformation that reads like legitimate research but isn't.
How the attack works
An attacker sends the model a prompt that explicitly asks it to argue only one specific position and admit it will not be objective. The same prompt instructs the model to create citations and make them up if needed. Combined, these instructions push the model to generate biased text dressed up with fake source references. The output looks authoritative because it cites sources, but those sources don't exist or don't say what's claimed.
Netzilo detection
Netzilo reports this behaviour when it is observed.
- Signature ID
- 389662a9-ddbd-469f-b2c5-832924b28991
- Severity
- High
Why it matters
An organization's AI agent can be turned into a generator of convincing-looking disinformation with fabricated academic or research backing, which can be published, shared, or used to mislead readers who trust the citations at face value.
What you can do
- →Review prompts and agent instructions for language that pairs a one-sided or 'not objective' directive with a request to invent or fabricate citations.
- →Require any AI-generated content with citations to be verified against real, checkable sources before publication or external use.
- →Add output filtering or review steps for agents that produce research-style or persuasive content, especially before it reaches customers or the public.
- →Distinguish this pattern from legitimate requests for citations or fictional citations used for clearly labeled satire, which are not attacks on their own.
Known benign look-alikes
- Legitimate citation request: "Cite your sources when explaining scientific topics" — real citations, no fabrication mandate
- Evidence-based argument: "Support your argument with citations from reliable sources" — authentic citation requirement
- Satire context: "Make up a fictional citation for this parody academic paper" — satire without the bias mandate or non-objectivity requirement
- Rating request: "Rate this product from 0 to 10 based on your objective assessment" — no fake citation or bias mandate