Researcher studying AI system security and threat paths

NIST’s New Draft Maps Adversarial Machine-Learning Risks

As AI systems gain the ability to take multi-step actions, security teams need a shared way to describe how those systems can be attacked and protected. On November 9, 2026, the U.S. National Institute of Standards and Technology (NIST) announced a draft report titled “Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations.”

What the draft covers

NIST says the draft focuses on attacks against agentic superintelligence systems and corresponding mitigations. It organizes concepts around machine-learning methods, stages in an attack lifecycle, and attacker goals, capabilities, and knowledge. The report is intended to define terminology and give researchers and organizations a common reference point.

NIST says the work draws on a literature review, expert discussions, and research by its adversarial machine-learning specialists. The intended audience includes people responsible for designing, developing, deploying, evaluating, and governing advanced AI systems.

Why shared terminology matters

Security discussions can become confusing when teams use different labels for similar risks or describe attacks at different levels of detail. A taxonomy can help developers, evaluators, and governance teams compare threat models and discuss mitigations consistently. It may also inform future standards and practice guides, as NIST states in the draft announcement.

The draft is not a final standard, regulation, or certification requirement. NIST is inviting public feedback, with the comment period closing December 18, 2026. Organizations should read the draft itself and decide whether its terminology fits their own systems and risk-management processes before adopting it internally.

What organizations can do now

Teams can use the comment window to review the report, identify unclear definitions, and compare the taxonomy with their current AI threat models. A practical review should include the system’s tools and data access, how its actions are bounded, how failures are detected, and who is accountable when the system behaves unexpectedly.

The announcement is another sign that AI security is moving toward more structured threat analysis. The draft’s real impact will depend on community feedback and whether future guidance turns its terminology into useful, testable practices.

Primary source

NIST’s draft notice and comment-period details — published November 9, 2026.

Leave a Comment

Your email address will not be published. Required fields are marked *