What Is NIKOLAI? CASRAI’s Frontier-AI-Safety Dictionary Explained

If you clicked through from a link about “NIKOLAI” and landed here wondering what you’re looking at: NIKOLAI is CASRAI’s own frontier-AI-safety dictionary — a single reference for the terminology that frontier AI labs, governments, and safety researchers each use slightly differently when they talk about capability thresholds, evaluations, safeguards, incident reporting, and who’s accountable for what. It is CASRAI’s own project. No lab, evaluator, or regulator has endorsed it, and it is not a standard in the way ISO or NIST publish standards — it’s an independent, unendorsed reference for looking a term up and seeing how it’s used across sources.

The name is an acronym: Nomenclature & Index for Knowledge Ontologies, Lexicons & AI Interoperability. In practice, people searching for it use much shorter phrases — “NIKOLAI AI safety dictionary,” “CASRAI NIKOLAI,” “frontier AI safety glossary,” “AI safety terminology dictionary” — and all of those land on the same thing, live at casrai.org/nikolai.

Why a dictionary like this exists

Frontier-AI-safety terminology didn’t develop in one place. Anthropic’s Responsible Scaling Policy, OpenAI’s Preparedness Framework, Google DeepMind’s Frontier Safety Framework, xAI’s Frontier AI Framework, and Meta’s Advanced AI Scaling Framework each define their own version of a “capability threshold,” a “safeguard,” and an “evaluation” — related concepts, described in each lab’s own words. Governments add their own layer on top: the NIST AI Risk Management Framework, the EU AI Act, California SB 53, and New York’s RAISE Act each use their own vocabulary for overlapping ideas, and voluntary international commitments — the Bletchley Declaration, the Seoul Summit frontier-AI-safety commitments — add a third layer again.

None of those sources cross-reference each other. A compliance team trying to map its own obligations against a lab’s public commitments, a researcher comparing how “incident” is defined across three frameworks, or a journalist trying to explain a safety-case argument to a general audience all run into the same problem: the same underlying idea, described in three or four incompatible vocabularies, with no shared index. NIKOLAI is CASRAI’s attempt at that index — not a new framework competing with the ones above, but a place to look up a concept and see how each source names and defines it.

What’s actually inside NIKOLAI

As of the current release, nikolai-v0.2, NIKOLAI defines 64 elements organized into 10 tracks, numbered N1 through N10:

• N1 — Actors, models and scope: developer, model identifier, deployment surface, coverage date, and the other scoping elements every framework has to define before anything else.
• N2 — Threat models and risk framing: threat model, risk pathway, and threat-actor profile — what a framework says it’s guarding against.
• N3 — Thresholds and checkpoints: capability threshold, alert threshold, halt condition, checkpoint rule — the operational core of a responsible scaling policy.
• N4 — Claims and argument: safety case, claim, risk level, residual risk determination — the structured argument that a model’s risk is acceptably low.
• N5 — Evidence and evaluations: evaluation, evaluation run, elicitation method, saturation status — the evidence a threshold or a claim actually rests on.
• N6 — Mitigations and security: safeguard, security level, security control, robustness level, monitor.
• N7 — Incidents: incident, incident type, discovery method, and incident reporting deadline and recipient — one of the few places this vocabulary has moved from voluntary framework into binding law (California SB 53, New York’s RAISE Act).
• N8 — Transparency and review: evaluator independence and conflict of interest, evaluator access attestation, publication rights clause, redaction.
• N9 — Commitments and governance: commitment, accountable decision-maker and sign-off, framework update protocol, roadmap item, and mapping declaration — the mechanism (explained below) for turning a NIKOLAI crosswalk row from a guess into a confirmed mapping.
• N10 — Assurance roles: declaring representative, CASRAI declaration reviewer, corroborating evaluator, declaration disputant — the accountability roles around NIKOLAI’s own Mapping Declaration process. Unlike N1–N9, which crosswalk real, named frontier-AI developer and regulator documents, N10 is CASRAI’s own editorial synthesis, structurally modeled on the CRediT contributor-role taxonomy (ANSI/NISO Z39.104-2022) that CASRAI also stewards — it is not something submitted to or adopted by any standards body.

How the site is organized

Every track has its own hub page at casrai.org/nikolai/track/{slug} — for example /nikolai/track/n3-thresholds-checkpoints for thresholds and checkpoints, or /nikolai/track/n7-incidents for incident reporting. Each of the 64 elements has its own definition page at casrai.org/nikolai/element/{slug}, such as /nikolai/element/safety-case or /nikolai/element/capability-threshold, showing the element’s operational definition plus its crosswalk — how each source framework defines or names the same concept.

Mapping Declarations — and the rule you need to know before you trust a crosswalk row

Every element page includes a crosswalk table showing how that concept maps across labs, regulators, and voluntary commitments. The critical rule: every crosswalk row NIKOLAI publishes is a “shadow mapping” — CASRAI’s own research-based reading of how a source uses a term — unless the organization itself has explicitly confirmed it. NIKOLAI is not an evaluator, an auditor, a regulator, or a lab consortium, and a shadow mapping is CASRAI’s best editorial judgment, not that organization’s own statement.

A Mapping Declaration, filed at casrai.org/nikolai/declare, is how that changes. It’s the mechanism a named organization — a lab, a regulator, an evaluator — uses to verify its own identity and then confirm, dispute, or add detail to how NIKOLAI has crosswalked its framework. Once a CASRAI editor reviews and approves a declaration, that specific row stops being a shadow mapping and is flagged as declared, with the declaring organization’s own words attached. Until that happens for a given row, treat it as CASRAI’s own interpretation, not as something the source organization has said about itself.

Programmatic access: API and MCP tools

NIKOLAI isn’t only a set of web pages. CASRAI publishes a public v1 REST API — documented at /implement/api-reference, with an OpenAPI 3.1 specification — that includes endpoints to list NIKOLAI’s ten tracks, list or fetch individual elements with their full crosswalk, and list published Mapping Declarations. Reads are keyless; a free bearer key raises the rate limit.

For agentic and LLM workflows, CASRAI also runs an MCP (Model Context Protocol) server, documented at /implement/mcp, with two NIKOLAI-specific tools: casrai_search_nikolai, which searches across all 64 elements by free text or track, and casrai_get_nikolai_element, which returns one element’s full definition and crosswalk by slug. Both are free and require no API key — and both tools’ own descriptions repeat the same shadow-mapping rule above, so an assistant surfacing a NIKOLAI result to a person knows not to present an unconfirmed crosswalk row as an organization’s own statement.

What NIKOLAI is — and isn’t

To be explicit, since this is the point most worth getting right: NIKOLAI is CASRAI’s own reference project. It is not an official standard, it is not endorsed by any AI lab, government, or standards body, and a crosswalk row appearing on a NIKOLAI page is not, by itself, evidence of what a lab or regulator actually said — check whether that row carries a declared Mapping Declaration before treating it as authoritative. What NIKOLAI is: CASRAI’s structured, versioned attempt to give compliance teams, researchers, and journalists one place to start when the same AI-governance concept shows up under three different names across a lab’s safety framework, a government’s AI regulation, and a voluntary international commitment.

Frequently asked questions

What does NIKOLAI stand for?

Nomenclature & Index for Knowledge Ontologies, Lexicons & AI Interoperability. It’s CASRAI’s own frontier-AI-safety dictionary of elements, not an acronym used by any lab or regulator.

Is NIKOLAI an official AI safety standard?

No. NIKOLAI is CASRAI’s own independent, unendorsed reference. No lab, evaluator, or regulator has endorsed it, and it should never be read as “the standard” for frontier-AI-safety terminology — it’s a place to compare how different sources use related terms.

How many elements and tracks does NIKOLAI have?

As of nikolai-v0.2, NIKOLAI defines 64 elements across 10 tracks (N1 through N10), covering scope and actors, threat models, thresholds, safety-case argumentation, evaluations, safeguards, incidents, transparency and review, commitments and governance, and NIKOLAI’s own assurance roles.

What is a “shadow mapping” in NIKOLAI?

A shadow mapping is any crosswalk row that reflects CASRAI’s own research-based reading of how a source defines a term, rather than that organization’s own confirmed statement. Every row is a shadow mapping by default unless a CASRAI-editor-approved Mapping Declaration has changed it.

What is a Mapping Declaration and who can file one?

A Mapping Declaration, filed at casrai.org/nikolai/declare, is how a named organization — an AI lab, a regulator, an independent evaluator — verifies its own identity and then confirms, disputes, or adds detail to a specific NIKOLAI crosswalk row for its own framework. Once approved by a CASRAI editor, that row is no longer a shadow mapping.

Can I access NIKOLAI programmatically?

Yes. CASRAI’s public v1 REST API (documented at /implement/api-reference) exposes NIKOLAI’s tracks, elements, and published Mapping Declarations, and CASRAI’s MCP server (/implement/mcp) offers two free, keyless tools — casrai_search_nikolai and casrai_get_nikolai_element — for agents and LLM tools to query it directly.

Browse NIKOLAI by track

Start with the track closest to what you’re trying to look up:

• N1 — Actors, models and scope
• N2 — Threat models and risk framing
• N3 — Thresholds and checkpoints
• N4 — Claims and argument
• N5 — Evidence and evaluations
• N6 — Mitigations and security
• N7 — Incidents
• N8 — Transparency and review
• N9 — Commitments and governance
• N10 — Assurance roles

Or see the full index, changelog, and versioning policy at casrai.org/nikolai.