NIST Just Handed Security Teams a Script for Their Own AI Prompts

What NIST actually shipped for nist compliance teams#

The fox does not chase every rustle in the hedgerow. It waits for the one worth the sprint, and on 19 August 2026, NIST gave the nist compliance world exactly that: a draft publication, SP 1353, titled the QuickStart Guide for Using Artificial Intelligence (AI) for Cybersecurity Framework (CSF) Analysis and Reporting. It is the first time NIST has formally written down how a practitioner should prompt an AI system to help with nist compliance work, rather than leaving every security team to invent its own shaky shortcut.

The document is short, practical, and structured around three worked scenarios rather than abstract theory. It walks a reader through using AI to run a governance review against CSF outcomes, to draft a Current State Profile from a pile of policies and interview notes, and to draft a Target State Profile from a set of mission objectives and regulatory requirements. Each scenario ships with sample prompts a team can lift directly, the kind of scaffolding that turns a vague intention into a repeatable task.

The comment clock is already running#

Public comment closes 15 October 2026 at 11:59 PM, addressed to [email protected]. That is a narrow window for a document that will likely shape how auditors, consultants and in-house GRC teams talk about nist csf 2.0 for years, and it is worth noting plainly: as of this week, no independent trade outlet has covered SP 1353 itself. The story is one day old. The first public flag came from a practitioner account on X, not a headline, which tells you something about how quietly government guidance can arrive even when it matters.

That single line, read the use cases, is the whole story in five words. NIST did not publish a manifesto about AI and security. It published a worked example, and a worked example is what a marketing team selling nist csf 2.0 services can actually build a campaign around, rather than another thinkpiece nobody asked for.

Read the draft closely and the restraint is the point. NIST does not tell a reader which model to use, does not promise a certified output, and does not claim the prompts replace a human reviewer. It hands over language, tested phrasing that turns a fuzzy instruction like check our CSF alignment into something an AI system can act on consistently, twice, three times, with the same reviewer getting comparable results each time. That consistency is the actual product. A prompt library that produces a different answer every time it is run is not a compliance tool, it is a slot machine wearing a lanyard.

It is also worth being honest about scope. SP 1353 covers three tasks, not the whole of nist compliance practice. It says nothing about incident response playbooks, nothing about vendor risk scoring, nothing about the harder judgement calls a senior GRC lead still has to make alone. What it does is take the three most repetitive, most administratively heavy tasks on that lead's desk and hand them a documented shortcut, which frees the hours that judgement actually needs.

Why the workforce cannot wait for permission#

Ask any GRC lead why they reached for a chatbot before NIST said it was fine, and the honest answer is a hedgerow full of thorns: too much framework, too few hands. The 2025 ISC2 Cybersecurity Workforce Study, a survey of 16,029 practitioners across four regions, found that 59% of organisations now report a critical or significant skills gap, up from 44% the year before. Ninety-five percent report at least one gap of any size, and AI itself is the single most-cited shortfall, named by 41% of respondents.

That last figure is the one that matters for nist csf 2.0 work specifically. A team that is already short-staffed and already reaching for AI on its own, without a shared vocabulary or a shared prompt library, is a team quietly building inconsistency into its own audit trail. Two analysts asking the same model two different questions about the same control will get two different answers, and neither is wrong, they are simply unrepeatable.

The pressure is not hypothetical. Thoropass's 2026 State of Audit and Compliance Report, drawn from more than 500 security, IT and compliance professionals, found that 69% already believe AI adoption is outpacing their own governance controls. Thoropass CEO Sam Li put it plainly: AI has moved faster than governance. That gap between adoption and oversight is precisely the gap SP 1353 is built to narrow.

Fifty-five percent of that same Thoropass sample name AI-related data exposure as their top breach concern, not a competitor's exploit, not a phishing campaign, their own tooling. A structured, NIST-endorsed prompt library does not remove that risk. It gives a security team a documented, repeatable way to point at when a client, an auditor or a board member asks how the AI is actually being used, which is a very different conversation from hoping nobody asks.

This is where most compliance vendors have been quietly caught out. A firm can sell nist compliance advisory work for years without ever writing down how its own analysts use AI day to day, because until 19 August there was no federal document to write it down against. That excuse is now gone. A prospect who has read the ISC2 numbers above will ask a very specific question in the next sales call: not whether you use AI, but which of NIST's three worked scenarios your workflow actually follows, and what happens when it does not.

The honest answer, for most firms selling nist compliance advisory work, is currently none of them, because the guidance is five days old at the time of writing. That is not a weakness to hide. It is a narrow, genuine window: the firms that can say we tested SP 1353 against a real engagement and here is what we found before the comment period closes will own that conversation for the rest of the year. The ones who wait for a competitor to publish first will spend Q4 catching up to a script someone else wrote.

The three prompts NIST actually wrote#

SP 1353 is deliberately narrow. It does not try to automate nist csf 2.0 compliance end to end, which is exactly why it is usable. It hands a practitioner three specific, worked scenarios, each with sample language a team can adapt rather than invent from a blank page.

Notice what is missing from that list: nothing about incident response, nothing about detection engineering, nothing glamorous. NIST picked the three tasks every GRC team already does badly under time pressure, the unglamorous paperwork that decides whether an audit goes smoothly or drags. That is a deliberate, sensible scent to follow.

Six functions, one shared vocabulary#

The guidance sits on top of the same six functions the wider nist cybersecurity framework 2.0 has used since its February 2024 launch: Govern, added new in version 2.0, then Identify, Protect, Detect, Respond and Recover. NIST Director Laurie Locascio called the CSF a vital tool for organisations of every size at that launch, and SP 1353 is the first document to lay AI-assisted prompts across all six rather than treating AI as a bolt-on extra.

KPMG's own regulatory alert on the draft makes the sharper point: even as voluntary guidance, SP 1353 may end up serving as a benchmark for how regulators judge cybersecurity diligence, precisely because it is the first document to say, in writing, what a defensible AI prompt for nist csf 2.0 work actually looks like.

That division of labour matters for how a firm actually staffs nist compliance work. A governance review sits squarely inside Govern, the function NIST added in version 2.0 specifically because earlier CSF releases treated oversight as an afterthought. A Current State Profile leans on Identify, the function every mature security programme already does reasonably well. The Target State Profile is the hardest of the three because it borrows from both, which is exactly why NIST gave it the most detailed sample prompt of the document.

The signals sitting either side of the draft#

SP 1353 did not land in a quiet field. NIST has been circling the same problem from several angles at once, and the pattern is worth reading as a whole rather than as one isolated draft. In December 2025, NIST opened comment on a companion effort, the draft Cyber AI Profile (NIST IR 8596), which attracted more than 6,500 people to its community of interest. Author Barbara Cuthill framed it simply: organisations need cybersecurity strategies that acknowledge where AI actually is, not where anyone wishes it were.

Cybersecurity Dive's coverage of that profile split it into three focus areas that echo neatly into SP 1353: securing AI systems themselves, using AI for defence, and blocking AI-powered attacks. SP 1353 is the practical, prompt-level layer sitting underneath the middle of those three, the part a working analyst touches on a Tuesday afternoon rather than the strategic document a CISO presents to the board.

None of this makes SP 1353 mandatory. Nothing about nist csf 2.0 has ever been mandatory in the way a licence or a statute is; it is a voluntary framework that happens to be the one 68% of practitioners already rate as the most useful thing on their desk. Voluntary and ignorable are not the same word, and treating them as synonyms is how a compliance vendor gets outflanked by a rival who read the use cases first.

Put the three signals side by side and a pattern emerges that no single press release states outright. NIST is not treating AI as one more control to bolt onto an existing checklist. It is rebuilding the checklist's own scaffolding, one worked example at a time, around the assumption that a human and a model will draft the paperwork together. CISA's alignment, the Cyber AI Profile's 6,500-strong community, and SP 1353's own three scenarios are three separate documents pointing at the same destination, which is rare enough in federal guidance to be worth naming plainly.

Fixing your nist compliance story before 15 October#

A fox does not announce the den it has already found. It moves in quietly, then defends the entrance. That is the right posture for any GRC vendor or security marketing team reading SP 1353 this week: build the muscle now, talk about it once the comment window closes and the guidance firms up.

Gartner's own August 2026 poll of 743 audit professionals found only 30% currently use AI for audit testing and just 12% for QA reviews, with Gartner analyst James Bourke noting that most current AI use concentrates on modest productivity gains rather than strategic audit work. That is the gap SP 1353's worked scenarios are aimed at closing, and it is a gap most competitors have not yet noticed is closable.

Submitting a comment is itself a marketing act, not just a civic one. A firm that engages with NIST's own process before 15 October has something genuinely new to say about nist compliance in Q4, rather than another recycled explainer of the six functions everyone has already written. Read the use cases, run them against a real client engagement, and note where the sample prompts fall short of a real audit trail. That gap is the actual content.

The nist csf 2.0 controls list itself has not changed. What changed is permission, in the loosest sense NIST ever grants it: a documented, citable way to say an AI-assisted nist compliance workflow follows a federal guide rather than a marketing team's improvisation. For any firm selling GRC or security advisory services, that is the difference between telling a prospect we use AI and showing them exactly which government document that use traces back to.

There is a second, quieter audience for this shift: the buyer side. A procurement team evaluating nist compliance vendors can now ask a direct, checkable question in an RFP, whether the vendor's AI-assisted workflow maps to SP 1353's three scenarios, and if not, why not. That single question turns a vague AI claim on a vendor's homepage into something a procurement lead can actually score. Firms that cannot answer it plainly will feel that gap first in the RFPs they lose, not in a headline.

Budget the next eight weeks accordingly. Between now and 15 October, the useful work is internal: test the prompts, document what breaks, decide which of the three scenarios your firm can genuinely stand behind. The external work, the case study, the comment letter, the sales deck slide, comes after, built on evidence rather than a rushed reaction to a five-day-old draft.

If your team wants the AI-assisted GRC workflow built rather than described, that is the discipline folkfox's cybersecurity marketing work is built around, and it is the same instinct behind our wider content marketing services: find the primary document before the trade press does, and say something true about it first.

Frequently asked questions#