What is a general-purpose AI model?

A general-purpose AI model is an AI model trained on large amounts of data that can perform a wide range of distinct tasks and can be built into many different downstream systems. Large language models are the familiar example. Under the EU AI Act, providers of these models have their own obligations, which have applied since 2 August 2025.

Why does it matter for AI agents?

Almost every AI agent is built on a general-purpose model supplied by a model provider. The agent adds a brief, sources, tools and bounds; the model supplies the reasoning and language. That split matters legally and operationally. The model provider has duties for the model. The firm that builds or deploys the agent has duties for the system and its use. Neither set replaces the other.

It also matters for control. You do not choose how the model was trained, and its behaviour can change when the provider updates it. Everything you need to govern the agent therefore has to sit outside the model.

What does the EU AI Act require of model providers?

In substance, providers of general-purpose AI models must:

• keep technical documentation of the model;
• provide information and documentation to the downstream providers who build systems on it, so they can understand its capabilities and limits and meet their own obligations;
• put in place a policy to comply with EU copyright law;
• publish a summary of the content used for training.

Models judged to carry systemic risk face further duties, including evaluation, risk mitigation, incident reporting and cybersecurity. Whether these rules reach firms that build agents on top of a model is covered in whether general-purpose AI model rules apply to agent builders. Usually they do not, unless the firm significantly modifies the model itself.

What does an example look like?

A specialist insurer builds a claims agent on a model from a large model provider. The model provider carries the general-purpose model obligations: documentation, information for downstream providers, the copyright policy and the training content summary. The insurer is the provider of its claims agent, as an AI system, and its deployer. It uses the model provider's documentation to understand the model's limits, then tests the agent on its own claims and sets its bounds. When the model provider releases a new version, the insurer treats it as a change and retests. What the insurer can and cannot rely on from the model provider is set out in what deployers inherit from general-purpose AI model obligations.

How is a general-purpose AI model different from an AI system?

A model is a component. An AI system is a working whole that takes inputs and produces outputs such as predictions, content, recommendations or decisions. An agent is an AI system: the model plus its instructions, sources, tools and the software around them. The Act regulates both, separately. High-risk classification applies to systems, according to their use; the general-purpose model obligations apply to models, whatever they end up being used for.

That is why a model provider's compliance tells you little about whether your agent is compliant. The model may be well documented and the agent still be high-risk, poorly overseen or acting outside its role.

Which terms sit close to it?

Foundation model and frontier model are informal terms for similar things. Provider and deployer are the roles in the Act. High-risk AI system is the classification that applies to agents by use. For the wider picture, see our guide to AI agent regulation.

This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.