A safety method that tests AI with attacks to find risks early.
An AI Red Team is like a kid pressing every “Do Not Push” button. If the robot burps secrets, you fix it before launch.
Teams use it before an AI goes live. They hunt for jailbreaks, leaks, and unfair answers.
Jailbreak
AI Red Teams use jailbreak prompts to test if a model breaks rules.
Prompt injection
Prompt injection is a common attack path for red teams to simulate.
Alignment
Failures found by red teams help improve Alignment.
AI Bug Bounty
AI Bug Bounty programs extend red team testing to outside communities.