Vision AI gets harder to trick with split attacks
A new defense stops attacks that hide malicious intent by splitting it across text and image inputs.
A new defense stops attacks that hide malicious intent by splitting it across text and image inputs.