Endorsed Compliance Framework Reduces Burden

General-purpose AI (GPAI) model providers demonstrate AI Act compliance via this voluntary code, drafted by independent experts and published July 10, 2025. The European Commission and AI Board endorse it as adequate, offering legal certainty over alternative methods. Signing cuts administrative load by providing concrete steps for Articles 53 (transparency, copyright for all GPAI) and 55 (safety/security for systemic-risk models). Sign via form to EU-AIOFFICE-CODE-SIGNATURES@ec.europa.eu; Signatory Taskforce (chaired by AI Office) ensures coherent rollout per Vademecum rules.

Targeted Chapters Deliver Practical Tools

Transparency (Article 53): Use the Model Documentation Form (DOCX) to log required info on model details, training data summary, and capabilities—streamlines disclosure without custom processes.

Copyright (Article 53): Implement a policy addressing EU copyright compliance, including opt-out mechanisms for training data and detection of rightholder-reserved content.

Safety and Security (Article 55, systemic-risk only): Apply state-of-the-art risk management—technical mitigations, red-teaming, adversarial testing—to curb risks like misuse or unintended harms from advanced models. xAI signed only this chapter, handling others separately.

Download chapters as PDFs for full practices.

Major Providers Commit Early

28+ signatories include Amazon, Anthropic, Cohere, Google, IBM, Microsoft, Mistral AI, OpenAI—proving viability for production-scale GPAI. List updates ongoing; taskforce fosters shared implementation.