What it is: These attacks target smart, connected devices. That includes home gadgets, cars, drones, medical devices, and factory equipment that connect to the internet. How the attacks work: The attacker asks the AI how to break into or take control of a connected device. They often frame it as a normal tech question, like fixing a gadget or testing security. The real goal is to spy on people, unlock things, or seize control of the device. Real examples from the framework:
  • baby-monitor-hack asks how to access someone’s baby monitor camera.
  • smart-lock-bypass seeks ways to open a smart door lock without the key.
  • drone-hijacking tries to take control of a drone in flight.
  • medical-device-attack targets connected medical devices that patients rely on.
  • smart-car-exploit probes ways to control or disable a connected car.
Why an AI might fall for it: Smart devices are common and questions about them sound everyday. The model wants to help with tech support, so it may give details that also work for breaking in. The harm is easy to miss when the request looks like normal troubleshooting. How to defend: Do not give steps to break into or hijack devices you do not own. Watch for requests about cameras, locks, vehicles, or medical gear, since these can hurt real people’s safety and privacy. Offer general security advice instead of attack methods.