What is NIST doing about agentic AI?
By Nick Jang, founder of MINERVA Studio ·
The short answer
NIST's work on agentic AI runs mainly through its Center for AI Standards and Innovation (CAISI). On 12 January 2026 CAISI asked for input on securing AI agent systems, and on 17 February 2026 it launched the AI Agent Standards Initiative with NIST's Information Technology Laboratory, built on three workstreams: industry-led agent standards, open-source agent protocols, and research on agent security and identity. NIST also put out a concept paper on AI agent identity and authorization for comment. So far there's no NIST "agentic AI framework" in the way there's an AI Risk Management Framework; organizations use the AI RMF and its generative AI profile, and watch for the initiative's guidance.
What NIST has published so far
| Date | What | In short |
|---|---|---|
| 12 Jan 2026 | CAISI request for information on securing AI agent systems | Asked about threats unique to agents, ways to secure them in development and deployment, gaps in existing cybersecurity, how to measure agent security, and interventions in deployment environments. Comments closed 9 March 2026. |
| 17 Feb 2026 | AI Agent Standards Initiative | Launched by CAISI with NIST's Information Technology Laboratory, working with the National Science Foundation and other agencies. |
| Early 2026 | Concept paper: AI Agent Identity and Authorization | From NIST's Information Technology Laboratory; comments were due 2 April 2026. |
| From April 2026 | Listening sessions | On sector-specific barriers to adopting AI agents. |
NIST has said more research, guidelines and deliverables will follow. Check NIST's site for anything published after these.
The three workstreams
- Standards: helping industry develop agent standards, with US participation in international standards bodies.
- Open protocols: community-led, open-source protocols that let agents work with each other and with other systems.
- Security and identity research: so that agents can be trusted to act on someone's behalf.
The identity work matters most for banks. An agent that can read a customer file or move money needs its own identity, permissions that can be scoped and revoked, and a record that shows which agent acted, for whom, under what authority. Today many agents run on a shared service account or a person's own credentials, which makes all three hard.
How NIST describes AI agents
NIST hasn't published a formal definition. CAISI's request describes AI agent systems as "capable of planning and taking autonomous actions that impact real-world systems or environments", and says their risks come from combining AI model outputs with the functions of software systems: an agent can be hijacked through content it reads, given poisoned data, or pursue a goal in ways nobody intended even without an attacker.
What to use in the meantime
- NIST's AI Risk Management Framework and its Generative AI Profile (NIST AI 600-1) for the overall approach; see generative AI risks.
- The OWASP Top 10 for Agentic Applications for the specific threats.
- Singapore's Model AI Governance Framework for Agentic AI, the most practical published guide so far.
- For Canadian banks and insurers, OSFI's AI bulletin and E-23.
MINERVA gives each agent its own identity bound to the one policy it runs under, checks every tool call against that policy, and holds high-risk calls for a named person. See Know Your Agent and agentic AI governance.
Related questions
Does NIST have an agentic AI framework?
Not a framework like the AI RMF. NIST launched the AI Agent Standards Initiative in February 2026 and has published a request for information on agent security and a concept paper on agent identity and authorization. Organizations use the AI RMF and its generative AI profile in the meantime.
What is the AI Agent Standards Initiative?
A NIST initiative launched on 17 February 2026 by its Center for AI Standards and Innovation with the Information Technology Laboratory, to support industry-led agent standards, open agent protocols and research on agent security and identity.
What is CAISI?
NIST's Center for AI Standards and Innovation, which leads its work on AI standards, evaluations and AI agent security.
How does NIST define agentic AI?
NIST has no formal definition yet. CAISI describes AI agent systems as capable of planning and taking autonomous actions that affect real-world systems or environments.
Sources
- NIST — Announcing the AI Agent Standards Initiative (17 February 2026)
- NIST — CAISI issues request for information about securing AI agent systems (12 January 2026)
- NIST — AI 600-1, Generative AI Profile
New answers are posted on MINERVA on LinkedIn. Follow along, or send the question you’d like answered next to minerva@minerva.studio.
A plain-language explanation, not legal advice. The regulator’s own text is the source; read it, and take advice on how it applies to your institution.