Report
svpino flags Fable 5.2’s 0%-versus-10% attack success gap across different tests
PromptArmor says FrontierBreach tests enterprise AI agents against attack chains involving ransomware, data theft and phishing.
TLDR
A spotless benchmark isn’t a force field. According to svpino, Palo Alto Networks reported a 0% attack success rate for Fable 5.2, while PromptArmor got 10% using its own attacks. These were different tests. His warning: blocking every attack in one evaluation doesn’t mean a model is secure.
Combined views
3.5K
4 Sources, first seen ago

Comments