EU

EU AI Office issued draft guidelines for obligations on general-purpose AI (GPAI) models applicable from August 2025. Stakeholders can provide feedback until May 22, 2025. The guidelines clarify the AI Act's provisions for GPAI, defining it as models performing multiple tasks, needing technical documentation and copyright compliance. Systems exceeding 10^25 FLOPs qualify as GPAI with systemic risk (GPAI-SR) and have stricter requirements. Fine-tuning these models may create new compliance obligations. Companies should establish AI governance, map AI applications, and prepare for the upcoming regulations. Compliance for earlier models must be achieved by August 2027.

https://www.wsgr.com/en/insights/eu-ai-office-clarifies-key-obligations-for-ai-models-becoming-applicable-in-august.html

EU Commission may intervene if technical standards for AI Act are delayed, currently set for 2026. Alternatives could aid compliance, though standards are not mandatory. The AI Act aims to regulate high-risk AI systems, with full enforcement expected by 2027.

https://www.euronews.com/next/2025/04/28/eu-commission-ready-to-step-in-if-ai-standards-delayed

EU Commission establishes guidelines for prohibited AI practices under AI Act, effective February 2025. Prohibitions include harmful manipulation, exploitation of vulnerabilities, social scoring, predictive criminal assessments, untargeted facial data scraping, emotion recognition, biometric categorization, and real-time remote biometric identification. Guidelines aim to clarify compliance and foster uniform application of the Act across the EU, though they are non-binding. Providers and deployers are responsible for ensuring AI systems meet regulations.

https://www.orrick.com/en/Insights/2025/04/EU-Commission-Publishes-Guidelines-on-the-Prohibited-AI-Practices-under-the-AI-Act

EU Commission clarifies AI definition: The Commission published guidelines detailing the definition of AI systems under the AI Act, outlining seven components, including machine-based systems, autonomy, adaptability, objective-driven outputs, inference capability, environmental interaction, and influence over environments. The guidelines help companies assess AI Act applicability. However, the guidelines are non-binding and not yet formally adopted.

https://www.orrick.com/en/Insights/2025/04/EU-Commission-Clarifies-Definition-of-AI-Systems

EU Commission invites feedback to clarify rules on general-purpose AI (GPAI) in a consultation aimed at shaping comprehensive guidelines to enhance understanding of GPAI concepts and compliance processes ahead of August 2025. Stakeholders can provide input until May 22, 2025.

https://digital-strategy.ec.europa.eu/en/news/commission-seeks-input-clarify-rules-general-purpose-ai-models

The European Commission's guidelines for the AI Act mandate transparency in training data for general-purpose AI models, requiring public summaries detailing the data used. This has sparked debate over what constitutes “sufficiently detailed” information, balancing rights holders' needs for access against providers' interests in protecting strategic assets. A template to aid in compliance was released in January 2025, structured into sections covering model details, data sources, and processing aspects, with final guidelines expected by mid-2025. Legal disputes will ultimately shape the regulation's implementation and future AI standards globally.

https://www.advant-nctm.com/en/news/the-european-commissions-template-on-training-data-transparency-first-guidelines-for-the-ai-act