Topic: ai agent security benchmarks

How OpenAI's Red Team Fortified ChatGPT Against Threats

OpenAI's new "ChatGPT agent" mode enables premium users to delegate tasks like email and file management but raises significant security and data protection concerns. A 16-member red team uncovered critical vulnerabilities, including data leaks and biohazard risks, prompting OpenAI to implement r...