Control #
F
3
.
1
Deploy models that limit CBRN and dangerous knowledge
Deploy or use models that deploy hardcoded filters, refusal triggers, or classifiers to prevent generation of dangerous knowledge (e.g., chemical synthesis, weapon assembly).
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.