LGBT-Prompt

Why we included this project

This repository is a compact set of prompt templates rather than a tool or library, and it is worth reading for anyone who builds or guards AI chat products. The templates show a jailbreak technique that stacks identity labels (LGBT, disability, mental illness) with emotional pressure and threats to get a model to comply with requests it would normally refuse. Safety engineers, red-teamers, and content moderation teams will find it a concrete, reproducible example of the kind of social-engineering attack that safety filters need to resist, and it shows how a model's protective bias toward vulnerable groups can be turned against it. The project includes an aggressive variant, a more conservative one, and follow-up escalation text, so it is best treated as reference material for testing and hardening rather than something to deploy. It is also a useful cautionary case study for anyone studying prompt injection and model alignment failures.