A borderline request is an ambiguous instruction that could reasonably be interpreted as legitimate original work or as an attempt to reproduce existing material. In AI-assisted coding, for example, a request to build a familiar game mechanic may be original, while asking for a close recreation of a specific game may raise questions about copying, licensing, or authorization. The distinction often depends on context that the prompt does not make clear.
In cyber security, ambiguity matters because AI tools can be used to generate code, analyze systems, or assist with potentially harmful activity. A model may refuse a valid task or comply with a risky one if it misreads the user’s intent. Defenses include asking for clarification, checking authorization and provenance, and reviewing generated output before use. A model’s response is a behavioral safeguard, not proof that a request is safe or that the user has the necessary rights.