• Researchers demonstrate that misleading text in the real-world environment can hijack the decision-making of embodied AI systems without hacking their software.
  • Self-driving cars, autonomous robots and drones, and other AI systems that use cameras may be vulnerable to these attacks.
  • The study presents the first academic exploration of environmental indirect prompt injection attacks against embodied AI systems.

Photos

you are viewing a single comment's thread
view the rest of the comments
[–] 1 point 9 months ago

Misleading text in the physical world can hijack AI-enabled robots, cybersecurity study shows

Eh, yes? Hasn't that been like that since day 1?

  • source