Defending MLLMs from Implicit Jailbreak Attacks
A new class of attacks where text and image look safe separately, but their combination carries malicious meaning
A new class of attacks where text and image look safe separately, but their combination carries malicious meaning