WHAT THE ARTICLE IS ABOUT
Voluntary but officially recognised compliance frameworks for GPAI providers
Article 56 creates a mechanism for codes of practice — collaboratively developed industry-regulator frameworks that GPAI providers can use to demonstrate compliance with their obligations under Articles 53 and 55. The AI Office facilitates and approves these codes, which carry regulatory weight even though adherence is technically voluntary.
WHAT IT SAYS
AI Office facilitates, industry develops, Commission can mandate if needed
- The AI Office facilitates the development of codes of practice to contribute to the implementation of Articles 53 and 55 — covering GPAI provider obligations
- Codes of practice must cover at minimum: how GPAI model technical documentation is kept up to date, the appropriate level of detail for training data summaries, identification and management of systemic risks, and measures for assessing and mitigating those risks
- The AI Office and AI Board ensure codes cover the required issues and invite GPAI model providers, civil society, independent experts, and other stakeholders to participate in their development
- Adherence to an approved code of practice is treated as demonstrating compliance with the covered obligations — until harmonised standards are published
- If no adequate code of practice is finalised by 2 August 2025 or the AI Office deems existing codes inadequate, the Commission may issue implementing acts establishing common rules for Articles 53 and 55
- A provider who neither adheres to a code nor complies with harmonised standards must demonstrate alternative adequate means of compliance to the AI Office
WHO IS AFFECTED
GPAI providers, the AI Office, and the broader AI ecosystem
- All GPAI model providers — codes of practice are their primary compliance pathway until harmonised standards exist
- The AI Office which facilitates and approves the codes
- Civil society organisations, independent experts and other stakeholders invited to participate in code development
- The European Commission which retains the power to issue implementing acts if codes are inadequate
- Downstream providers and deployers who benefit from transparency about which providers adhere to which codes
WHAT IT MEANS FOR SMES
A practical compliance signal when choosing AI vendors
- The GPAI Code of Practice was published in 2025 — if you use AI models from companies like OpenAI, Anthropic, Google or Meta, check whether those companies have signed up to and are complying with the code
- Adherence to an approved code of practice is the most visible signal of a GPAI provider’s regulatory compliance status — it is publicly documented and the AI Office tracks it
- For SMEs building AI products on top of GPAI models: knowing whether your foundation model provider adheres to the code affects your own compliance position
- The code covers training data transparency — which has direct implications for copyright and the legality of AI-generated content you produce using those models
- Monitor AI Office announcements on code of practice approvals and compliance — this is live regulatory intelligence relevant to every business using foundation AI models
Related Articles
- Article 53 — GPAI baseline obligations (codes of practice demonstrate compliance with these)
- Article 55 — Systemic risk obligations (also covered by codes of practice)
- Article 40 — Harmonised standards (the mechanism that will eventually supersede codes of practice)
- Article 64 — AI Office (the body that facilitates and approves the codes)
