Red Teaming

A structured approach where dedicated teams or individuals attempt to find vulnerabilities, biases and failure modes in AI systems through simulated attacks and adversarial testing.

Hiring people to deliberately try to break or trick your AI before bad actors do. Like a bank hiring someone to test their vault security by attempting a (controlled) break-in.

Red teaming is a best practice for AI governance that helps organisations discover risks before they cause harm. Incorporating regular red teaming into your AI lifecycle demonstrates proactive risk management to regulators and stakeholders.