On December 16, 2025, the U.S. Nationwide Institute of Requirements and Expertise (“NIST”) printed a preliminary draft of the Cybersecurity Framework Profile for Synthetic Intelligence (“Cyber AI Profile” or “Profile”). Based on the draft, the Cyber AI Profile is meant to “present tips for managing cybersecurity danger associated to AI programs [and] determine[] alternatives for utilizing AI to boost cybersecurity capabilities.” The draft Profile makes use of the prevailing voluntary NIST Cybersecurity Framework (“CSF”) 2.0 — which “gives steering to business, authorities businesses, and different organizations to handle cybersecurity dangers” — and overlays three AI Focus Areas (Safe, Detect, Thwart) on high of the CSF’s outcomes (Features, Classes, and Subcategories) to recommend issues for organizations to prioritize when securing AI implementations, utilizing AI to boost cybersecurity defenses, or defending towards adversarial makes use of of AI. This draft steering will possible be acquainted to organizations that already leverage the CSF 2.0 of their cybersecurity applications and could be complimentary to present frameworks that organizations have already got in place. Even so, the outcomes are designed to be versatile such {that a} vary of organizations (with mature or novel applications) can leverage the steering to assist handle AI-related cybersecurity danger.
For entities or stakeholders that could be excited about providing suggestions on the preliminary draft, NIST is planning to host a workshop on January 14, 2026, to debate the draft. The Profile can be open for remark till January 30, 2026. Under, we briefly summarize the Profile’s organizational construction, in addition to areas on which NIST is looking for public remark.
Focus Areas
The Cyber AI Profile is organized into three Focus Areas that handle AI-related cybersecurity danger from completely different however overlapping angles.
- Safe – “[F]ocuses on managing cybersecurity challenges when” organizations combine an AI system into their setting. Examples of the usage of AI that fall inside the scope of Safe embody the usage of AI by “[p]ower grids to stability masses” and “[c]ustomer service organizations to carry out preliminary interactions with prospects.”
- Defend – Goals to determine alternatives for the makes use of of AI that help cybersecurity processes and actions. For instance, AI can improve cyber protection capabilities associated to mission assurance, proactive danger administration, predictive upkeep and danger forecasting, “[a]dvanced risk detection and evaluation,” adversarial coaching and simulation, and automatic incident response.
- Thwart – Emphasizes constructing resilience to guard towards AI-enabled threats. For instance, AI-enabled spear-phishing assaults exploit customers by means of extra lifelike manipulation utilizing deepfakes and generative AI. These AI-enabled assaults assist stress the necessity to replace coaching for personnel and to have automated defenses that bolster safety measures.
Cybersecurity Framework 2.0 Core
As indicated above, the Cyber AI Profile is split into the six CSF features: Govern, Establish, Defend, Detect, Reply, and Get better. Inside every CSF perform, the Cyber AI Profile gives pattern focus space issues. Just like the CSF 2.0, the Cyber AI Profile doesn’t provide particular directions for easy methods to obtain the advisable outcomes however as an alternative gives references for the way a corporation may take into account reaching the advisable outcomes — instance informative references embody sources corresponding to technical papers, the European Union Company for Cybersecurity’s (“ENISA”) Menace Panorama 2025 paper, NIST Particular Publication (“SP”) 800-218, and Databricks AI Safety Framework Model 2.0. The Cyber AI Profile additionally gives advisable priorities to assist decide which Subcategory organizations ought to prioritize: “1” for Excessive Precedence, “2” for Reasonable Precedence, and “3” for Foundational Precedence.
Govern
The Govern profile captures the institution, communication, and monitoring of “[t]he group’s cybersecurity danger administration technique, expectations, and coverage.” This profile is additional divided into the CSF classes of (1) group context, (2) danger administration technique, (3) roles, tasks, and authorities, (4) coverage, (5) oversight, and (6) cybersecurity provide chain danger administration. AI issues below these classes embody, for instance:
- Safe/Defend: Figuring out and speaking AI dependencies.
- Safe/Defend: “[I]ntegrat[ing] AI-specific dangers into the group’s formal danger urge for food and tolerance statements.”
- Defend/Thwart: Growing AI-specific risk info sharing channels.
- Defend: Augmenting cybersecurity groups with AI brokers.
- Defend: Utilizing AI to conduct governance checks to determine operational conflicts.
- Defend: Conducting risk detection of supplier-provided AI fashions.
Establish
The Establish profile focuses on guaranteeing that “[t]he group’s cybersecurity dangers are understood.” This profile consists of the CSF classes of (1) asset administration, (2) danger evaluation, and (3) enchancment. AI issues below these classes embody, for instance:
- Defend: Inventorying software program and programs to incorporate “AI fashions, APIs, keys, brokers, information . . . and their integrations and permissions.”
- Defend/Thwart: Incorporating AI-specific assaults as a part of one’s vulnerability administration program.
- Defend: “Defin[ing] circumstances for disabling AI autonomy throughout danger response.”
- Defend: Integrating “AI-specific procedures for containment” throughout incident response.
Defend
The Defend profile serves as a framework to make sure that “[s]afeguards to handle the group’s cybersecurity dangers are used” and is split into sections targeted on (1) figuring out administration, authentication, and entry management, (2) consciousness and coaching, (3) information safety, (4) platform safety, and (5) know-how infrastructure resilience. AI issues below these classes embody, for instance:
- Safe: Issuing AI programs distinctive identities and credentials.
- Safe/Defend/Thwart: Growing AI-related consciousness and coaching for personnel.
- Safe/Defend: “Keep[ing] protected, often take a look at backups of crucial AI belongings.”
- Safe: Limiting the execution of arbitrary code by AI agent programs.
- Defend: “Implement[ing] AI-specific resilience mechanisms.”
Detect
The Detect profile goals to supply a regular to make sure that “[p]ossible cybersecurity assaults and compromises are discovered and analyzed.” This profile is split up into two classes: (1) steady monitoring, and (2) opposed occasion evaluation. AI issues below these classes embody, for instance:
- Defend: “[F]lagging anomalies, correlating suspicious behaviors, and recognizing uncommon patterns sooner than people and different automated instruments.”
- Thwart: “Personnel could also be topic to AI-enabled phishing or deepfake assaults.”
- Thwart: “AI-enabled cyber assaults may determine and exploit” vulnerabilities launched by “[t]hird-part[ies] . . . [through] updates and/or patches to software program and programs.”
- Safe: Figuring out what “new monitoring is required to trace actions taken by AI.”
Reply
The Reply profile gives steering to verify “[a]ctions relating to a detected cybersecurity incident are taken” and consists of the CSF classes of (1) incident administration, (2) incident evaluation, (3) incident response reporting and communication, and (4) incident mitigation. AI issues below these classes embody, for instance:
- Safe: “Set up[ing] standards for triaging and validating AI-related incidents.”
- Defend: “Integrat[ing] AI-driven analytics into incident categorization and prioritization to determine and flag AI-influenced occasions.”
- Safe: Diagnosing complicated assaults with new instruments and strategies.
- Thwart: “[S]earch[ing] for indicators of adversary AI utilization within the incident.”
Get better
The Get better profile covers restoring “[a]ssets and operations affected by a cybersecurity incident” and consists of (1) incident restoration plan execution, and (2) incident restoration communication. AI issues below these classes embody, for instance:
- Defend: Utilizing AI to “speed up[] restoration by calculating which programs to revive first, monitor[] progress, and draft[] clear updates to maintain stakeholders knowledgeable.”
- Defend: Utilizing AI to “forecast {hardware} failures and system degradation.”
- Defend: Evaluating “how AI protection programs carried out” after an incident.
Whereas in preliminary draft kind, the Cyber AI Profile joins a rising checklist of AI-related steering such because the NIST AI Danger Administration Framework in addition to different steering that’s below improvement such because the NIST SP 800-53 Management Overlays for Securing AI Programs (“COSAiS”).
Request for Public Remark
As famous above, NIST is accepting feedback on the draft till January 30, 2026, along with planning a workshop on January 14, 2026, to debate the draft. The preliminary draft states that NIST is particularly looking for public touch upon the draft within the following areas:
- Doc construction and subjects:
- How do you envision utilizing this publication? What modifications would you wish to see to extend/enhance that use?
- How do you count on this publication to affect your future practices and processes?
- Are the proposed subjects on this doc adequate to assist your group prioritize cybersecurity outcomes for AI?
- Focus Space descriptions (Part 2.1):
- How properly do the Focus Space descriptions mirror the scope and traits of AI utilization? Are any traits lacking, and if that’s the case, what are they and the way ought to we describe them?
- Profile content material (Sections 2.3–2.8):
- When excited about making use of the Cyber AI Profile, how helpful (or not) is it for all three Focus Areas to be proven alongside one another (as they’re presently mirrored)? What worth may there be in offering Profile content material for every Focus Space individually?
- What format(s) could be helpful for offering the knowledge within the Cyber AI Profile (e.g., a spreadsheet/workbook, the NIST Cybersecurity and Privateness Reference Instrument (CPRT))?
- How properly do the priorities and issues mentioned in Sections 2.3–2.8 relate to present practices and requirements leveraged by your group? Are there vital gaps between present practices and people which can be needed to deal with distinctive traits of AI in every Focus Space that this publication ought to handle? How ought to the AI-specific issues inform the prioritization of every Subcategory?
- NIST printed the Cybersecurity Framework (CSF) 2.0 Informative References and Implementation Examples to point out potential methods to attain the end result in every Subcategory. This preliminary draft consists of examples of Informative References for the Cyber AI Profile. Additional literature evaluate is in progress and NIST is looking for extra enter on Informative References to incorporate. Which further AI cybersecurity tips, requirements, greatest practices, or mappings are you utilizing that you simply suggest including as Informative References for the Cyber AI Profile? For any Informative References you suggest, please share with us why you suggest them in addition to how and why you’ll prioritize them for this doc.
- Glossary (Appendix B):
- NIST welcomes requests and ideas for phrases that ought to be added to this doc’s Glossary