title: "L’IA est son propre meilleur garde-fou : Glean utilise l’IA pour détecter les tentatives de jailbreak avec une précision de 97,8 %"

Glean utilizes a specialized AI model to detect and prevent adversarial attacks, such as jailbreaking and prompt injection, on large language models. This detection model achieves a 97.8% accuracy rate in identifying malicious prompts designed to bypass safety features. This capability is part of a multi-layered security approach that also includes robust permission checks, data loss prevention, and hallucination detection. The overall strategy aims to provide comprehensive protection for enterprises deploying generative AI, ensuring that the models behave as intended and company data remains secure.