AI Act Obligations for General-Purpose AI Models Apply from 2 August 2025

From 2 August 2025, the obligations for providers of general-purpose AI models (GPAI) under the EU AI Act become applicable. If your organisation develops, trains, or makes available foundation models in the EU (whether via API, cloud, download, or integration) the compliance clock starts now.

This post explains:

• Who must comply;

• What obligations apply;

• What happens with models already on the market;

• Key guidance and templates from the European Commission.

What is a General-Purpose AI Model (GPAI Model)?

According to the AI Act, a general-purpose AI model (GPAI model) is "an AI model that is trained with a large amount of data using self-supervision at scale, is capable to competently perform a wide range of distinct tasks, and that can be integrated into a variety of downstream systems or applications." (AI Act, Article 3(44))

The Commission further explains in its Guidelines that GPAI models are:

• Typically foundation models capable of tasks beyond a single pre-defined purpose;

• Model-level assets (not full applications) that serve as the base for building other AI systems;

• Not limited to language models - they can also include multimodal, code generation, vision, or speech models;

• Often made available via:

  • Download;

  • API;

  • Open-source release;

  • Cloud service integration;

  • Embedded in hardware or applications.

The obligations in Chapter V of the AI Act apply at the model level, regardless of whether the model is provided as a standalone tool or integrated downstream into other systems.

The AI Act distinguishes between AI models and AI systems, and the provider of the model is responsible for Chapter V obligations even if others integrate or fine-tune the model later.

What obligations apply under Chapter V?

Under Chapter V of the AI Act, the following obligations apply from 2 August 2025 to GPAI models placed on the EU market on or after 2 August 2025 (per Article 111(3) of the AI Act):

• Maintaining technical documentation about the model;

• Sharing documentation with downstream AI system providers;

• Implementing a copyright compliance policy;

• Publishing a public summary of training data content;

• Appointing an EU representative if the provider is not established in the EU;

• For models with systemic risk (e.g. exceeding 10²⁵ FLOP or designated by the Commission):

  • Conducting model evaluations;

  • Performing systemic risk assessments;

  • Reporting serious incidents;

  • Implementing cybersecurity safeguards.

According to Article 111(3) of the AI Act, providers of general-purpose AI models placed on the market before 2 August 2025 have until 2 August 2027 to bring those models into compliance with all obligations under Chapter V.

For example:

• ChatGPT‑4o, which is already on the EU market before 2 August 2025, benefits from this transitional regime.

• A future ChatGPT‑5o, placed on or after 2 August 2025, must comply fully and immediately.

This distinction is essential for determining whether obligations like technical documentation or training data summaries apply now or later.

Support for providers preparing to release general purpose AI models

The Commission recognises that some providers are currently training or finalising models expected to be placed on the market in the coming weeks or months.

In such cases, especially where the model may exceed the systemic risk threshold, the Commission encourages early and proactive engagement with the AI Office.

Key implementation resources from the European Commission

To assist providers, the Commission has published several key documents:

1. Guidelines on the scope of obligations for GPAI providers - includes definitions, scope, thresholds, and transitional provisions (see Section 5.3 for market placement timing);

2. FAQ on the GPAI guidelines;

3. Template for the public summary of training data content;

4. FAQ on the template above.

Final thoughts

The entry into application of Chapter V of the AI Act marks a significant regulatory shift for developers and providers of general-purpose AI models in the EU. If you are planning to release or update a foundation model, or to integrate a foundation model into an AI system, it's high time to review your legal position and compliance obligations.

Early compliance planning and documentation will be essential to avoid delays and ensure responsible deployment.