Skip to content
Misbehavior1 sourceFirst seen
Picture from the cnn.com article
Image: cnn.com

Hack the hardware, and the chatbot starts giving very bad advice

Ourtake

Securing the model is not enough if the computers running it can be quietly rewritten underneath it.

What happened

Security firm Eclypsium showed that AI hardware, including a Nvidia computing platform, can be remotely tampered with to alter a model's answers. In demonstrations, a chatbot advised a user with chest pain to stay home instead of going to hospital. Nvidia said the scenarios were not previously unknown to the industry and that memory protections can block the attack path.

Source