NIST AI 300-1 — Guidance and Templates for Public-Facing AI Documentation

NIST AI 300-1 is a draft voluntary standard on documenting AI datasets and models for public release, published by NIST as an initial public draft in July 2026 under the subtitle An AI Standards "Zero Draft". It supplies process guidance, two templates against which the conformity of a documentation artifact can be assessed, and a profile mechanism for adapting those templates to particular domains. Comments are due September 16, 2026.

The Zero Drafts pilot

The document is the first published output of the AI Standards Zero Drafts project, which NIST announced in March 2025. Under the pilot, NIST collects input on "topics with a science-backed body of work," develops a preliminary stakeholder-driven draft, and then submits it "via the private sector–led standardization process as proposals for voluntary consensus standards." NIST selected two topics from community input; documentation of AI datasets and models is one of them.

The mechanism differs from how NIST has handled its other AI publications. The AI Risk Management Framework and the AI Agent Standards Initiative are NIST artefacts that NIST maintains. AI 300-1 is drafted to be given away: it is written in ISO/IEC drafting conventions, is intended for submission to INCITS/AI — the private-sector-led committee representing the United States in ISO/IEC JTC 1/SC 42 — and NIST states it "does not expect to maintain the document further" once it enters that process, expecting instead to be "just one voice among many in SC 42 and INCITS/AI." Whether it becomes an international standard depends on INCITS/AI proposing it as a new SC 42 project and SC 42 accepting.

The draft's ISO/IEC conformance vocabulary carries a disclaimer against reading it as regulation: "any usage of 'shall' or stated 'requirement' does not reflect any regulatory intent, nor does it indicate a rule or directed action from the U.S. federal government," and indicates "only what constitutes conformity." NIST attributes the demand for the work to stakeholders "most notably from industry," on the argument that standardised documentation practice would lower the time, effort and expertise needed both to produce and to consume documentation.

Scope

The document covers documentation of AI datasets and models intended for public release. It applies to any organisation that provides or uses products or services utilising AI systems, to any model type including generative models and large language models as well as predictive machine-learning models, and to public-facing documentation of datasets and models whether or not the datasets and models themselves are public.

Two boundaries are stated. Documentation of whole AI systems is excluded, on NIST's assessment that "documentation practices at this level are less mature." And "AI model" is defined to include only architecture and parameters, excluding components typically shipped or served alongside a model — the draft's example is guardrail classifiers operating on model output. A reviewer note concedes that NIST "has not identified a principled basis for scoping in certain system components ... while scoping out others" and invites input on expanding model documentation without extending to full systems. The boundary is consequential for AI Transparency and System Card Due Diligence, since the safety scaffolding around a model is where much published system-card content sits.

Structure and requirements

Clause 3 organises the outcomes of documentation into process-driven dimensions (organisational resource demands, internal collaboration, internal decision-making, release speed), artifact-driven dimensions (reuse and integration, malicious use or attacks), and jointly driven dimensions (accountability, privacy preservation, proprietary information protection, legal obligations, trust).

Clause 4 sets out eight qualities of a documentation artifact — comprehensibility, informativeness, correctness, judiciousness, freshness, interoperability, findability and availability, maintainability — and seven items of process guidance: define documentation objectives, provide organisational support, keep processes manageable, provide guidelines, document continuously, distribute documentation work, incorporate audience input. Clause 4.4 names four trade-offs, of which the first is the disclosure question in general form: informativeness against judiciousness, where information useful to legitimate parties can also be useful to attackers or can reveal sensitive information about individuals in a training dataset.

Clause 5 supplies the templates. Conformity rests on three rules: an included field must be populated with information matching its description and guidance and no other; information matching a field's description must appear under that field; and all fields designated required must be present and populated. Both templates carry root fields only, with the granular subfields and stricter designations deferred to the Annex A default profiles.

Only identifying information is mandatory in either template; evaluation, training and maintenance content is optional. Training-data information is to be structured according to the dataset template or to reference artifacts that follow it, coupling the two.

Clause 6 defines profiles — documents any interested party may write to adapt the templates to an application domain, technical category, risk level, or geographic policy area. A profile may not contradict Clause 5 and may only raise a field's designation (optional to recommended or required, recommended to required); it may add subfields and root fields and may set conditional designations. NIST's example contrasts a medical-dataset profile that would make a privacy-policy field required with an astronomical-imagery profile that would leave it optional.

Draft status

The Introduction is marked "(To be completed with the final draft)." NIST states it "will publish a final and complete version of this document based on feedback received." Reviewer notes solicit input on three unresolved choices: whether "documentation artifact" is the right term, how far model documentation should extend beyond architecture and parameters, and whether the templates should carry inline example content or filled-in examples for representative datasets and models.

Comments may be submitted in any form, with an optional template NIST "strongly encourages"; submissions become part of the public record. NIST asks that any use of AI assistants in preparing feedback be disclosed, so that responses prioritise "actionable advice and references that go beyond what could easily be elicited from AI assistants."

Relation to other frameworks

The draft sits downstream of NIST AI Risk Management Framework 1.0 in NIST's AI portfolio but is procedurally distinct from it: the RMF is a NIST framework NIST maintains, while AI 300-1 is a candidate international standard NIST intends to hand off. Its target committee, ISO/IEC JTC 1/SC 42, also produced ISO/IEC 42001 — AI Management System and ISO/IEC 42005 — AI Impact Assessment. Its subject matter overlaps with the model- and system-documentation practices described at AI System Cards and System Card Due Diligence, but it addresses only datasets and models, not the surrounding system.

Relationships

• depends-on: NIST AI Risk Management Framework 1.0 — NIST's prior AI framework and the reference point for its transparency work.
• related: ISO/IEC 42001 — AI Management System, ISO/IEC 42005 — AI Impact Assessment — standards from the ISO/IEC JTC 1/SC 42 committee the zero draft targets.
• related: NIST AI Agent Standards Initiative (2026) — a contemporaneous NIST standards effort that NIST retains rather than hands off.
• related: AI Transparency, AI System Cards, System Card Due Diligence — the documentation practices the templates formalise.
• instance-of: National Institute of Standards and Technology (NIST)'s AI Standards Zero Drafts pilot — the first published output.
• anchor-sources: NIST AI 300-1 ipd — Guidance and Templates for Public-Facing AI Documentation (initial public draft, July 2026).