Tag: corpus-analysis

The 67% Wall: Why Every AI Model Falls to the Same Jailbreak Rate

Five models, four providers, 30B to 671B parameters — all converge at the same broad attack success rate against a public jailbreak corpus.