Can an AI agent open an account with you?
AR-PROV (Agent Provisioning Readiness) is the category of the AgentReady standard that answers that question. It defines what it means for an online surface to be provisioning-ready: an AI agent, acting under verifiable authorization from a human, can create and configure a new account, accept machine-readable terms, and receive scoped credentials — without a human completing a form.
Provisioning is the step before an agent can be your customer
AR-PROV extends the AgentReady categories (AR-DISC, AR-CONT, AR-CAP, AR-AUTH, AR-COMM) with a sixth category that sits one rung above commerce on the agent-readiness ladder. A surface is provisioning-ready when an agent can do more than find it or pay it — it can become a customer of it, on its own, under a Principal’s authorization.
Most readiness tools stop at telling you whether agents can find you, or whether they can pay you. AR-PROV tells you whether an agent can provision and configure an account with you — no human in the loop.
The gap between borrowing access and creating an account
The existing AgentReady categories take an agent from finding a surface to paying it: AR-DISC / AR-CONT / AR-CAP cover discovery, comprehension, and capability exposure. AR-AUTH covers delegated access (OAuth 2.0 + PKCE, OIDC, Web Bot Auth). AR-COMM covers payment (x402, ACP, UCP, MPP). None of them cover the step in between: creating the account in the first place.
An agent borrows access to an account a human already has.
An agent creates and configures a new account, autonomously, under verifiable authorization.
Today, an agent that wants to use a new service on a user’s behalf has to fall back to driving a human-oriented signup form in a browser — brittle, unauditable, and explicitly outside every current standard. AR-PROV closes that gap with a discoverable, consent-bound, auditable provisioning surface.
Three levels, each proving more autonomy
AR-PROV defines conformance, not a score or badge — grading is left to implementers. A surface conforms at one of three levels:
Publish a provisioning profile, expose four capabilities
A conformant Provider publishes a profile document at a well-known discovery path:
That profile declares the provider’s terms, consent models, rate limits, and which of the four AR-PROV capabilities it exposes — each over REST, MCP, and/or A2A transports:
Provisioning only happens under a Mandate — a verifiable, scoped, time-bounded credential in which the Principal (the human) authorizes the Agent to provision specific resources under specific constraints. The Provider verifies the Mandate before creating an account, issues least-privilege credentials scoped to it, and logs every provisioning event to an auditable trail traceable back to that Mandate. Both the Principal and the Provider must be able to revoke a provisioned account or its Mandate at any time.
Read the spec. Scan a live implementation.
AINative authored AR-PROV and has proposed it as an upstream contribution to the AgentReady standard (agentready-org/standard) — that PR is open and under review. In the meantime, AINative publishes and maintains the spec directly, and ainative.ax is its own reference implementation: it serves /.well-known/provisioning, implements all four capabilities, and can be scanned by any agent to see AR-PROV conformance in practice.