On December 22, 2025, OpenAI announced a security update for the ChatGPT Atlas browser agent aimed at reducing the risk of prompt injection attacks. The update follows internal red teaming that identified new exploit classes, leading to the deployment of an adversarially trained model designed to better resist malicious instructions embedded in web content. OpenAI describes this update as part of a continuous "proactive rapid response loop" to address evolving security challenges in agentic browsing.