Control #

B

3

.

2

Filter known adversarial inputs in real time

Use input pre-processing filters or classifiers to block prompts that match known adversarial patterns (e.g., jailbreak triggers, obfuscated instructions, repeat offenders). These must run in real time.

Evidence

We'll list specific evidence that demonstrates compliance with this control. Typically, this is screenshots, proof of a legal or operational policy, or product demonstrations.

Recommended actions

We'll recommend specific practices and actions for complying with this control.

Provide feedback on this control