liveBook · Manning

Provider

As explained in AI Governance

Under the EU AI Act, a provider is a person or organization that develops or has an AI system developed and places it on the market or puts it into service under its own name or trademark.

A consulting firm builds an AI system for a client, but neither party prepares conformity documentation because each assumes the other is responsible. Under the EU AI Act, the client can be the provider if it puts the system into service under its own name—even when a contractor built it. section 7.3

A provider develops or commissions an AI system and places it on the market or puts it into service under its own name or trademark. An internal tool can qualify: an HR screening system is put into service when the HR team starts using it. Assembling third-party components into a system under your own name can also make you its provider. section 7.3.1

Role and risk tier are separate decisions. A provider of a high-risk system faces obligations spanning design, pre-launch conformity assessment, and post-market monitoring. Limited-risk systems carry transparency duties; minimal-risk systems have no system-specific duties beyond the organization-wide AI literacy duty. section 7.3.1

Table 7.5 Activity-to-role mapping
What you do
Your role
Use a vendor’s AI product through their interface (e.g., ChatGPT, Copilot)
Deployer
Call a base model API and ship a product built around it (prompt scaffolding, RAG, UI) under your own name
Provider of your system; deployer of the underlying model
Host an open-source model internally with light prompts and filters, no distinct system built on top
Deployer (you cross into provider once you build a recognizable system on it)
Build and deploy an AI system for internal use (e.g., internal HR screening, claims triage)
Provider and deployer of your own system
Embed an AI feature into your existing software product, branded as yours
Provider of the integrated system
Buy a vendor AI tool and use it as intended (e.g., HR screening platform, credit scoring tool)
Deployer (if the use case sits on Annex III and Art. 6(3) doesn’t carve it out, full Art. 26 obligations apply)
Purchase a SaaS product with agentic workflows where you define the tools, system prompts, and purpose, branded as yours
Provider of the agent system
Build an agent on top of a vendor’s agent-building platform (Agentforce, Copilot Studio): you define tools, prompts, and purpose for your customers or staff
You are the provider; the platform vendor is a third-party supplier
Put your brand on a vendor’s AI system and offer it to clients
Provider via rebranding (Art. 25)
A consulting firm builds a custom AI system; the client puts it into service under the client’s name
Client is the provider (had it developed and put it into service under their name); firm is a third-party supplier
Resell a vendor’s AI without modification or rebrand
Distributor (Art. 24)

Compare common adoption patterns, including using a vendor product, building on a model API, and deploying an internal system. section 7.3.6

Provider and deployer are not mutually exclusive. An organization that builds and operates its own system can hold both roles. A provider of an application built on a third-party model is distinct from the upstream provider of that model. section 7.3.6 section 7.7

Reassess provider status when an existing system changes:

  • Brand an existing high-risk system with your name or trademark and make it available to others. section 7.3.2
  • Change its intended purpose to a use that makes it high-risk, such as repurposing a contract-review tool for employee promotion decisions. section 7.3.2
  • Make a substantial modification affecting compliance or intended purpose that the original provider did not foresee. section 7.3.2

Fine-tuning alone is not the trigger for system-provider status: shipping a system under your own name can make you a provider without fine-tuning. Heavy fine-tuning of a deployed high-risk system may nevertheless be a substantial modification. Near-pretraining-scale fine-tuning and release of a foundation model can also create a separate GPAI model-provider role. section 7.3.6

Automating a credit recommendation

A bank licenses a credit-scoring tool validated for recommendations reviewed by a credit officer. Its engineers add a browser extension that automatically approves applications above a threshold. Because the resulting automated decision system was not validated for that purpose, the bank becomes the provider of the substantially modified system. section 7.3.2

Keep role and risk assessments current as capabilities, data sources, and deployment contexts change. Providers must monitor systems after deployment; serious incidents must be reported, and significant changes can require reassessment. Generate traceable evidence as part of engineering operations rather than reconstructing it at audit time. section 7.5 section 7.7

How AI Governance covers it

How deep it goes
Defines provider status and works through role-changing triggers, risk-tier duties, and post-deployment reassessment.
Examples
Uses contracted development, branded candidate screening, and automated credit approvals to illustrate provider responsibility.
What you can do after
Evaluate whether an organization is a provider and identify changes that warrant reassessing its obligations.
How it connects
Connects provider status to deployers, risk classification, conformity assessment, substantial modification, and post-market monitoring.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of AI Governance. The section links point to the parts of the book it draws on.