What Happened

The EU AI Act is a comprehensive legislation on artificial intelligence that aims to foster responsible AI development and deployment. The regulation sets out rules that scale with the level of risk an AI system or model might pose, ranging from unacceptable risks (e.g., violating human rights) to minimal risks (e.g., systems that don't interact directly with people). For open-source developers, the regulation introduces new obligations, including providing clear documentation and adding tools to disclose model information when deployed. The EU AI Act applies broadly, not just to those within the EU. If an open-source developer outside the EU offers or impacts people within the EU, their AI systems or models are included in the Act. The regulation takes effect over the next two years, with obligations enforced starting August 2026.

Background and Context

The EU AI Act is a binding regulation that aims to preserve open research and support small and medium-sized enterprises (SMEs). To achieve this, it sets out rules for different levels of the AI stack, including providers, deployers, distributors, and users. The regulation works at different levels of risk, with obligations varying depending on the type of model or system. For general-purpose AI (GPAI) models, there is another risk category called systemic risk. GPAI models using substantial computing power (over 10^25 FLOPs for training) or having high-impact capabilities are subject to specific obligations. Currently, only eight models from seven developers meet the default systemic risk criterion.

Why it Matters to the Industry

The EU AI Act has significant implications for open-source developers in the adult industry. The regulation introduces new obligations, including providing clear documentation and adding tools to disclose model information when deployed. For limited-risk AI systems, developers need to disclose that users are interacting with an AI system, mark synthetic content as artificially generated or manipulated, and follow existing copyright and privacy rules. For open-source non-systemic risk GPAI models, developers must draft and make available a sufficiently detailed summary of the content used to train the model. They also need to implement a policy to comply with EU law on copyright and related rights, including respecting opt-outs from AI training. The regulation ties into existing regulations on copyright and personal data, such as the copyright directive and data protection regulation. Open-source developers can rely on tools like Hugging Face's integrated support for better opt-out mechanisms and personal data redaction to comply with these obligations.

What Comes Next

The EU AI Act is a binding regulation that will be enforced starting August 2026. To prepare for compliance, open-source developers should review the regulation and its implications on their projects. They can rely on tools like Hugging Face's integrated support for better opt-out mechanisms and personal data redaction to comply with these obligations. The European Commission will help create codes of practice with guidelines for detecting and labeling artificially generated content. These codes are currently being written with industry and civil society participation, and are expected to be published by May 2025. Open-source developers can get involved in shaping how the Act's provisions play out through public consultations and working groups.

Key Facts

  • The EU AI Act is a comprehensive legislation on artificial intelligence that aims to foster responsible AI development and deployment.
  • The regulation sets out rules that scale with the level of risk an AI system or model might pose, ranging from unacceptable risks to minimal risks.
  • For open-source developers, the regulation introduces new obligations, including providing clear documentation and adding tools to disclose model information when deployed.
  • The EU AI Act applies broadly, not just to those within the EU. If an open-source developer outside the EU offers or impacts people within the EU, their AI systems or models are included in the Act.
  • Obligations for GPAI models will be enforced starting August 2025.