It’s Frighteningly Easy to Jailbreak Some Frontier AI Models

Researchers tested a tool designed to bypass safety features in artificial intelligence models from four major companies and documented the effectiveness of the circumvention method.

Next story