Anthropic warns that refactoring attacks disguise harmful fragments as harmless fragments

Anthropic releases Constitutional Classifiers++: Combat Universal Jailbreaks with about 1% of the computing power overhead

On January 9, 2026, Anthropic released research articles and papers to launch "Next-generation Constitutional Classifiers" (also known as Constitution...