Don’t do SECURITY. Do business SECURELY.

NIST updates its adversarial machine learning taxonomy to cover AI agents

NIST’s draft SP 2400-102e2026 updates its taxonomy of attacks on AI systems, with a focus on AI agents; comments close on 18 December.

On 9 November 2026 the US National Institute of Standards and Technology (NIST) published an initial public draft of NIST SP 2400-102e2026, Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations. It updates the taxonomy last issued as NIST AI 100-2 E2025 in March 2025, with a new focus on attacks against AI agents.

Key points

  • The draft classifies attacks on AI systems, such as evasion, poisoning, privacy attacks, and prompt injection, and the mitigations for each.
  • It gives particular attention to agentic systems that take actions, use tools, and work with other agents.
  • NIST wants input on the latest attacks, mitigations that work in practice, and terminology that needs standardising.
  • Comments are open until 18 December 2026 and go to SP-2400-102@nist.gov.

The taxonomy is widely used as a common vocabulary in AI risk assessments, red teaming, and supplier questionnaires. Organisations deploying AI agents can use it to check that their ISO/IEC 42001 risk assessment and security testing cover these attack types, and those with practical experience may want to respond to the consultation.

Source: NIST Computer Security Resource Center

This update is general information, not legal advice. If you would like help assessing the impact on your information security or privacy programme, contact us.

Get our weekly insights by emailA free weekly threat and vulnerability round-up with a link to our threat log spreadsheet, plus legal, regulatory, standards, and AI updates.
Subscribe free

More insights