General-Purpose AI Models: New Obligations Under the EU AI Act

The EU AI Act introduces specific obligations for general-purpose AI models (GPAIs), including large language models and foundation models.

What Qualifies as General-Purpose AI?

A GPAI model is defined as an AI model that:

• Can be used for multiple purposes
• Can be integrated into various systems
• Has significant generality

Examples: GPT-4, Claude, Gemini, Stable Diffusion

Obligations by August 2025

All GPAI Providers Must:

1. Technical documentation: Comprehensive model documentation
2. Information to downstream providers: Usage instructions and limitations
3. Copyright compliance: Respect for copyright law
4. Content policy: Publication of detailed training content summary

Systemic Risk Models (>10^25 FLOPs)

Additional requirements:

• Model evaluation and adversarial testing
• Serious incident tracking and reporting
• Cybersecurity protections
• Energy efficiency reporting

Documentation Requirements

Technical Documentation Must Include:

• Model architecture and training process
• Training data sources and selection criteria
• Model capabilities and limitations
• Computational resources used
• Testing and validation results

Information for Integrators:

• Integration guidelines
• Intended uses and restrictions
• Performance benchmarks
• Safety measures required

Practical Implementation

1. Start documentation now - Retrofitting is difficult
2. Implement monitoring systems for incident tracking
3. Establish downstream communication channels
4. Prepare for audits and compliance checks

Open Source Considerations

Open source models have limited exemptions:

• Exempt from some documentation requirements
• Must still comply with copyright
• Systemic risk obligations still apply if >10^25 FLOPs

The August 2025 deadline is approaching fast. Start preparing your GPAI compliance strategy today.