What changed
NIST released draft SP 2400-102e2026, a taxonomy of adversarial-machine-learning attacks and mitigations focused on agentic systems. It organizes attack life-cycle stages, goals, capabilities and attacker knowledge, with an agentic attack index for fine-grained references.
Why it matters
A shared vocabulary can make threat models, test plans and incident records comparable across teams. Because the document is a public draft, practitioners can also shape terminology before later standards and practice guides build on it.
What to do
Map one current agent threat model to the draft taxonomy, note missing attack or mitigation classes with concrete evidence, and submit comments to NIST by 18 December 2026.
Caveat
This is an initial public draft, not a binding standard or a validated control catalog. NIST says it is intended to inform future work; teams still need system-specific testing and risk decisions.