The federal government’s AI security agenda in the new year should center on intent security, purple-teaming and guidance.