Industry Analysis 8 min read

GPAI Obligations Now Active: What Model Providers Must Know

August 2025 brought new obligations for General-Purpose AI models. Here's what the law requires, who must comply, and how the voluntary Code of Practice fits in.

By Thomas Klein, Privacy Counsel · Published

GPAI Rules Are Now Active

As of August 2, 2025, providers of General-Purpose AI (GPAI) models must comply with new EU AI Act obligations. This milestone affects foundation model providers worldwide who serve the EU market.

What Qualifies as a GPAI Model?

Under Article 51, GPAI models are defined by their general capability and training compute. GPAI Models are AI models with significant generality that can perform a wide range of distinct tasks, trained using large-scale self-supervision on broad data. Systemic Risk GPAI are models meeting the cumulative compute threshold (currently presumed at 10^25 FLOPs) or designated based on capability assessments.

Core Obligations for All GPAI Providers

  • Technical Documentation: Maintain and update comprehensive model documentation covering training, testing, and capabilities
  • Training Data Summary: Publish a public summary using the EU's standardized template (published July 2025)
  • Copyright Compliance: Respect EU Copyright Directive; implement processes to honor opt-out requests from rights holders
  • Instructions for Use: Provide downstream users with clear usage guidance and capability descriptions

Additional Requirements for Systemic Risk Models

  • Conduct and document adversarial testing ("red teaming") to identify systemic risks
  • Assess and mitigate risks throughout the model lifecycle
  • Report serious incidents to the AI Office within two weeks
  • Implement robust cybersecurity protections against model compromise
  • Notify the Commission within two weeks of meeting the systemic risk threshold

The Voluntary Code of Practice

The Commission published a voluntary GPAI Code of Practice in July 2025. While not legally mandatory, adherence provides a "presumption of compliance" with AI Act obligations—reducing regulatory burden and legal uncertainty.

Enforcement Timeline

  • August 2, 2025: GPAI obligations take effect for new models
  • August 2, 2026: EU Commission enforcement powers begin (fines, recalls, mandatory mitigations)
  • August 2, 2027: Legacy models must fully comply

What This Means for Enterprises

For organizations deploying GPAI-based applications, understanding your provider's compliance posture is increasingly important. When evaluating AI vendors, consider asking: Have they published a training data summary? Do they follow the Code of Practice? How do they handle copyright compliance? These questions help assess regulatory risk in your AI supply chain.