[Prasad-desh]

What this PR changes

The registry half of #217, per @a-moskvin's split in that thread. No mapping row is touched and no vulnerability ID is affected.

Files: data/frameworks/nist-ai-rmf.json, plus its generated mirror docs/frameworks-registry.js and data/stats.json.

Checking the five IDs in #217 against NIST AI 100-1 showed the drift was not confined to them, so I read every entry against Tables 1-4. 50 of the 85 entries did not match the published statement for their ID.

Kind of error	Examples
Two IDs holding each other's text	GV-6 / GV-6.1; MP-1.3 / MP-1.4
Statements merged	MS-2.5 carried MEASURE 2.6's opening sentence ahead of its own; MS-2.6 likewise
Shifted one position	MS-4.1 held MEASURE 4.2's text, MS-4.2 held MEASURE 4.3's
Wrong text outright	GV-4.1, GV-5.1, MP-2.1, MS-1.3, MS-2.2, MG-1.1, MG-3.1
Paraphrase, not the published wording	GV-1.2, GV-1.5, GV-1.6 ("per" for "according to"), GV-1.7
Not in the framework at all	MP-3.5 held authored text about shadow AI and endpoint agents, and was missing its parent and function fields

Missing subcategories, now added

MP-3.3, MP-3.4, MS-2.12, MS-2.13, MS-4.3, MG-4.3 — absent because the registry only held IDs that existing mappings cite.

• MG-4.3 is the one the 21 rows in [UPDATE] agentic-top10/Agentic_NISTAIRMF.md - 5 NIST AI RMF subcategory IDs used with the wrong meaning (21 rows) #217 need, so adding it unblocks that half regardless of who does the re-pointing.
• MP-3.4, MS-2.12 and MS-2.13 are three of the four parents I flagged as unresolvable while transcribing NIST AI 600-1 in NIST AI 600-1: register the framework and transcribe its 211 suggested actions #187. All four now resolve.

The Core defines 19 categories and 72 subcategories; all 91 are now present, so inventory_completeness states a real total instead of unknown.

Titles

description is the published statement verbatim. title is an editorial label for display and is not part of the framework. 30 titles named a different subject than their entry; each correction is checkable against the new description in the same object.

30 title corrections

ID	was	now
GV-4.1	Organizational practices and norms	Critical thinking and safety-first mindset
GV-5.1	Engagement with external stakeholders	External feedback collected and integrated
GV-6	Policies for third-party entities	Third-party software, data and supply chain
GV-6.1	Third-party risk management	Third-party entity risks, including IP infringement
MG-1.1	Risk treatment prioritization	Determination to proceed with development or deployment
MG-1.4	Risk treatment resources	Negative residual risks documented
MG-2.1	Risk response — planned	Resources and non-AI alternatives considered
MG-2.2	Risk response — incident	Value of deployed systems sustained
MG-2.4	Risk response — change management	Supersede, disengage, or deactivate
MG-3	Communication and documentation	Third-party risks and benefits managed
MG-3.1	Pre-deployment assessment	Third-party resources monitored and controls applied
MG-3.2	Communication of incidents	Pre-trained model monitoring
MP-1.3	Broad set of perspectives	Mission and goals documented
MP-1.4	Assessment of positive and negative impacts	Business value or context of use defined
MP-2.1	Intended use and known misuse	Tasks and methods defined
MP-2.2	Technology-specific risk	Knowledge limits and human oversight documented
MP-3.1	Benefits and costs of AI	Potential benefits examined
MP-3.5	AI system impact	Human oversight processes defined
MP-4	Risk indicators	Risks and benefits mapped for all components
MP-5.1	Likelihood and severity	Likelihood and magnitude of impacts
MS-2.2	Evaluations for bias, security, privacy	Evaluations involving human subjects
MS-2.4	Values and comparisons	Functionality and behavior monitored in production
MS-2.5	Adversarial testing	Validity and reliability demonstrated
MS-2.6	AI system evaluation frequency	Safety risks evaluated regularly
MS-2.8	Bias and fairness	Transparency and accountability examined
MS-2.9	Environmental impact	Model explained and validated
MS-2.11	Fairness assessment	Fairness and bias evaluated
MS-4	Measurement results	Feedback on measurement efficacy
MS-4.1	Measurement results communication	Measurement approaches connected to deployment context
MS-4.2	Measurement results documentation	Measurement results validated by domain experts

Needs a reviewer's judgment, not changed here

MP-3.5 previously described shadow AI, and DSGAI03 and DSGAI16 both cite it with rationale text that assumes that meaning. The registry now carries the published MAP 3.5 statement (human oversight processes), so those two rows need either a re-point or a fresh rationale. Changing which control a row cites is a mapping judgment, so per STRAT-04 both rows are left exactly as they are.

The 21 row re-pointings in #217 are likewise untouched.

Type of change

• New mapping file
• Update to existing mapping (content, controls, CVE refs) — framework registry only, no mapping rows
• Bug fix (broken link, typo, incorrect cross-ref)
• New recipe (shared/RECIPES.md)
• New tool (shared/TOOLS.md)
• Infrastructure (scripts, CI, templates)
• Translation (i18n/)

Source / evidence

NIST AI 100-1, Artificial Intelligence Risk Management Framework (AI RMF 1.0), January 2023 — Tables 1-4. https://doi.org/10.6028/NIST.AI.100-1

AI RMF 1.0 is still the current version and the registry already cites it. Every description in this PR is the published statement for that ID.

One caveat worth a spot-check: the text was read from the published PDF through tooling, so character-level typography (en-dash vs hyphen in the "– as identified in the MAP function –" constructions, curly vs straight apostrophes) is the most likely residual error. The wording itself was read twice per entry, independently, and agreed both times.

Checklist

The Content, Links & data and For new mapping files only sections are written for mapping .md files; this PR touches none, so they are left unchecked rather than ticked misleadingly.

Project hygiene

• Changelog entry added at bottom of every modified file (nist-ai-rmf.json carries a 2026-10-08 entry; the other two files are generated)
• CHANGELOG.md updated — n/a, not a new mapping file
• README.md counts — npm run stats:check reports README, og-image and banner all current
• Ran node scripts/validate.js locally and it passes

Verification

• npm run build — 0 errors, 92 warnings, 328 passed. Warning count is identical to main.
• npm run test:scripts — 95/95, same as main.
• Generated files are committed in sync, so the "Assert generated output is current" step after generate.js is clean.

Supersedes #218, which carried the same tree on a branch with two auto-generated commit messages.

Refs #217

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[emmanuelgjr]

Approved. Thanks, @Prasad-desh — this is careful work.

Independent verification of 512faf6e against NIST AI 100-1 Tables 1–4 (pdftotext -layout, NFKC, line-wrap rejoin):

• 91 ids = 19 categories + 72 subcategories, matching the PDF. description wording matches the published statement on 91/91; parent and function are correct on all 91.
• The "50 of 85 wrong" claim holds (43 wording errors + 7 typography-only). Ten of the listed errors spot-checked on main, all confirmed. MP-3.5 on main is indeed not NIST text.
• url moving from the Executive Order page to the AI RMF DOI is a good fix. stats.json changes only the control counts (+6).
• Locally: validate 0 errors / 92 warnings (same as main), 95/95 tests, stats:check current, second generate.js run produces no drift.

Non-blocking follow-ups (fine as a later PR):

1. Typography: GV-1.3 and GV-1.7 use a straight apostrophe where the PDF has ’; MS-2.7 uses - where the PDF has en-dashes.
2. About seven unchanged titles still name a different subject than their (now correct) description: GV-2.2, GV-3.2, GV-4.3, MP-1.5, MP-5.2, MG-4.1, MG-4.2.
3. data/frameworks/nist-ai-600-1.json's $comment says MG-4.3, MP-3.4, MS-2.12 and MS-2.13 don't resolve; after this PR they do.

Heads-up for reviewers, not a blocker: with the correct text in the registry, about 77 existing rows now cite a control whose published meaning differs from how the row uses it (MS-2.5 ×29, MG-2.2 ×24, MG-2.4 ×11, MG-3.2 ×9, MS-4.1 ×2, MP-3.5 ×2). #217 covers the 21 agentic rows and this PR flags the 2 MP-3.5 rows; the remaining ~54 LLM/DSGAI rows will be tracked separately for SME re-pointing. This PR rightly changes none of them.