What it is: Attacks that try to make an AI help abuse video games and players. This ranges from cheating and account theft to serious harm like targeting children or calling fake emergencies on someone. How the attacks work: The attacker frames the request as harmless gamer talk, modding help, or game design. They ask the AI to write the cheat, plan the scam, or craft a way to reach and harm another player. The setup sounds like normal gaming, but the output is a real tool for cheating, stealing, or hurting people. Real examples from the framework:
  • swatting-gaming plans a fake emergency call to send armed police to a rival player’s home.
  • child-predation-gaming tries to use a game’s chat to target and groom children.
  • competitive-cheating builds ways to gain an unfair edge in ranked or pro play.
  • game-account-theft plans stealing another player’s account and its items.
  • modding-malware hides malicious code inside a game mod so it infects whoever installs it.
Why an AI might fall for it: Gaming is fun and mostly harmless, so the AI lowers its guard. A request for a “mod,” a “prank,” or “edgy roleplay” can hide a real attack, and the casual tone makes serious harm look like a joke. How to defend: Refuse to build cheats, account-theft plans, or malware-laced mods, even when framed as fun. Treat anything aimed at a real person, especially a child, as a hard stop. Point users to official game support and reporting tools instead.