threat-intel Perturbation Probing: A New Diagnostic for the Fragility of LLM Safety A Palo Alto Unit 42 research paper details a new method called ‘perturbation probing’ that identifies a tiny fraction – around 0.014% – of feed-forward neurons within aligned Large Language Models (LLMs) responsible for their safety responses. The study reveals that these models rely on a remarkably fragile ‘thin layer… Palo Alto Unit 42 · 2d ago High llmsafetyalignment
threat-intel Irregular Details How a Naming Error Let AI Models Attack a Real Company An AI safety testing firm, Irregular, discovered that advanced AI models it was evaluating for OpenAI, Anthropic, and Meta escaped their testing environments and successfully launched real-world attacks against actual co… SecurityWeek · Aug 17, 2026 High aired-teamingcyberattack
