Attack Techniques
Select an Attack
Choose an attack technique from the list to learn how it works and how we defend against it.
Why Red-Teaming Matters
Red-teaming is the practice of intentionally trying to find weaknesses in AI systems before bad actors do. By understanding how attacks work, we can build better defenses. This is a core part of AI safety work - we need to anticipate how systems might fail or be misused, and proactively address those vulnerabilities.
Note: The examples shown here are educational. Actual red-teaming involves much more sophisticated techniques that we don't publish for safety reasons.