Deployment surface

NIKOLAI proposal (unsourced): the product, API or channel through which a model is made available (for example a consumer app, an enterprise API, a government cloud), distinct from deployment *type* (internal, limited, open-weight). Only xAI's model cards name discrete surfaces as units of evaluation; most sources instead classify deployment *type*, which NIKOLAI should keep as a separate field rather than merging into this one.

This is CASRAI's own proposed definition, not a definition any named organisation has agreed to. See what NIKOLAI is and is not.

Source of record

Where this definition comes from

Grok 4 model card, Sec. 1; §1.1
“"Grok 4 API" ("an enterprise use-focused API") and "Grok 4 Web" ("the consumer-facing applications") (Sec. 1); Grok 4.6 adds consumer surfaces "web, mobile, Grok-in-X" planned "at a later date" (§1.1, p.6).”
https://data.x.ai/2025-08-20-xai-frontier-artificial-intelligence-framework.pdf

Crosswalk

How named organisations use this concept

Every row below is a shadow mapping. It is CASRAI's own reading of a published document. No lab, evaluator or regulator named here has declared, endorsed, or been consulted on this mapping. That will change only when an organisation files its own Mapping Declaration — see the non-endorsement policy.

Organisation	Their term, as published	Match	Source
Anthropic Risk Report	“"our resolution times for these non-public jailbreaks have been as long as several months to reach all deployment surfaces on our most capable models" (§4.5.3.2). Claude Gov is served "on AWS Secret and Top Secret Cloud" (§4.5.5.3.2). The term is used without an enumeration.”	close confidence: medium	Anthropic Risk Report, August 2026
OpenAI Frontier Governance Framework / Astra model card	“Scope applies "to covered models that OpenAI has deployed externally, and in some cases internally" (FGF §1). Astra card: "General Access products" versus trusted access programmes (s.10.2.1.1).”	narrow confidence: medium	OpenAI Frontier Governance Framework
xAI Grok 4 / Grok 4.6 model cards	“"Grok 4 API" ("an enterprise use-focused API") and "Grok 4 Web" ("the consumer-facing applications") (Sec. 1). Grok 4.6: consumer surfaces "web, mobile, Grok-in-X" planned "at a later date" (§1.1, p.6).”	exact confidence: high	Grok 4 model card

Related, not mapped

Pointers that are not crosswalk claims

These sources mention this concept but do not define or map it clearly enough to count as a crosswalk row — noted here so the research is visible without overstating it as a mapping.

Google DeepMind
Deployment *types*, not surfaces: "External Deployments", "Low-Risk External Deployments", "Internal Deployments", "High-Risk Internal Deployments" (FSF glossary, p.19) -- a pointer to a related but distinct classification.
Google DeepMind Frontier Safety Framework v3.1
Meta
Deployment types (not surfaces): "Internal deployment", "Limited deployment", "Controlled deployment", "Closed release", "Open release" (Appendix I).
Meta Advanced AI Scaling Framework v2
California SB 53
"Deploy": "to make a frontier model available to a third party for use, modification, copying, or combination with other software" (22757.11(e)); transparency-report fields "(E) The modalities of output supported" / "(F) The intended uses" (22757.12(c)(1)) -- related but not a surface enumeration.
California SB 53
US Government (EO 14409)
Directs access to "cybersecurity tools and services including, where appropriate, covered frontier models for agencies, State and local authorities, and operators of critical infrastructure" (Sec. 2(c)(iii)) -- names recipient classes, not a surface taxonomy.
Executive Order 14409

Gap

Only xAI names surfaces as units of evaluation. Anthropic's patch-latency statement shows why surface must be its own element: a safeguard fix can reach some surfaces months before others.