Research finds Ai hardware hack

(CNN)– Researchers have uncovered a novel, and dangerous way to hack Ai models, go after the hardware.

They found that by hacking the machines that actually power Ai, they could tamper with the answers chatbots give.

Security experts are particularly concerned about the reliability of Ai models when there is no suggestion they have been tampered with.

Attacking the very hardware that powers them has the potential to alter them in a way that is more difficult to detect.