Research paper on 10,800 jailbreak attempts and non-linear features that predict jailbreak success.
February 19, 2026
November 2025 workshop paper introducing a dataset of 10,800 jailbreak attempts across 35 attack methods and analyzing non-linear features in prompts that predict jailbreak success. Free, advanced resource for understanding why certain jailbreak patterns work and how to design more principled red-team probes.