CoFiBreak preprint reports more diverse successful attacks on AI models

Researchers report in a preprint that CoFiBreak, a method for training AI models to test other models' safety defenses, discovered 1.6-2.0x as many distinct successful fine-grained attack strategies as the strongest baseline across frontier victims on HarmBench. The method rewards underused strategies and discourages repetitive presentations, applying diversity rewards only to attacks that exceed an effectiveness threshold. These rewards operate during training, avoiding added inference cost from reward computation. On AdvBench, attacks reached a 71.0% success rate with 10 generations per intent against Claude-Opus-4.6, an unseen target. The researchers say code will be released.