NIST's adversarial ML guidance: 6 action items for your security team

The report reviews NIST’s 2025 adversarial machine learning guidance as a strong foundational framework—providing standardized terminology, a comprehensive taxonomy across attack types, learning methods, and lifecycle stages—while emphasizing its current limitations against rapidly evolving AI threats. Industry experts underscore that traditional AST/SCA miss model-level and supply-chain risks (e.g., malicious pickle deserialization), urging organizations to inventory AI assets (including AI-/ML-BOMs), generate SBOMs that capture model and dataset provenance, apply binary analysis, secure CI/CD and training environments, align controls to NIST lifecycle stages, and implement AI-specific incident response, with heightened caution around agentic AI and models sourced from public repositories.