Definition: GPAI Code of Practice
The GPAI Code of Practice is a voluntary, Commission-published framework that translates the EU AI Act’s transparency, copyright, and safety obligations for general-purpose AI model providers into concrete, checkable commitments.
Core characteristics of the GPAI Code of Practice
The Code is soft law: it adds no new duties beyond the AI Act, but gives providers a shared template for proving they meet existing ones.
- Drafted by independent experts via an AI Office-led consultation
- Organized into three chapters with distinct obligations
- Signed per chapter, so providers can commit to some, not others
- Monitored by a Signatory Taskforce the AI Office chairs
GPAI Code of Practice vs. General-Purpose AI Model and EU AI Act
The EU AI Act is the binding regulation creating GPAI obligations. A General-Purpose AI Model is the legal category those obligations apply to. The Code is neither the law nor the category; it is one voluntary way a provider can show it meets the law’s requirements.
Importance of the GPAI Code of Practice in enterprise AI
Most enterprise AI runs on a model from a handful of large providers, so the Code’s signatory list works as a quick compliance signal. Bitkom’s 2026 AI study found a quarter of German companies had not engaged with the AI Act at all, making signatory status a practical shortcut.
Methods and procedures for the GPAI Code of Practice
Three steps turn a provider’s Code commitment into practice.
Signing and commitment process
A provider notifies the AI Office which chapters it adopts, then applies the commitments across releases via the Signatory Taskforce.
- Submit a signed declaration naming the adopted chapters
- Appoint an internal contact for AI Office dialogue
- Publish the documentation the chapter requires
Chapter-by-chapter compliance mapping
Transparency requires model documentation and a training-content summary; Copyright requires a policy respecting EU copyright law; Safety and Security, limited to models with systemic risk, requires risk assessment and incident reporting, per Articles 53 and 55 of the AI Act.
AI Office monitoring and taskforce dialogue
The AI Office does not audit signatories against a fixed checklist the way a formal conformity review does. It runs an ongoing dialogue through the Taskforce instead.
Important KPIs for the GPAI Code of Practice
Tracking Code adherence across a vendor portfolio needs a short scorecard.
Vendor signatory tracking
- Chapters signed per vendor: 0-3
- Technical documentation on file: yes or no
- Copyright policy published: yes or no
- Last Code-related update reviewed: date
Strategic exposure
Gartner projects more than 80% of enterprises will have used generative AI APIs by 2026, up from under 5% in 2023, concentrating most companies’ Code exposure in a small number of upstream vendors.
Documentation quality
A vendor’s technical documentation is the clearest proxy for how seriously it treats its commitments, stronger than a signatory badge alone.
Risk factors and controls for the GPAI Code of Practice
Treating the Code as a compliance guarantee creates its own risks.
Non-signatory and partial-signatory exposure
A provider that has not signed, or signed only part of the Code, still carries full legal obligations; it proves compliance a different way.
- Confirm which chapters, if any, a vendor has signed
- Request the underlying documentation, not just the status
- Flag vendors that change signatory status after procurement
Code drift from binding law
Because the Code is soft law, the AI Office can revise it without changing the AI Act, so today’s wording may already be outdated guidance.
False sense of conformity
Code adherence supports a presumption of compliance but is not conclusive proof. Deployers building high-risk systems on a signatory’s model still carry their own obligations.
Practical example
A 95-employee specialty coatings manufacturer in Rhineland-Palatinate was evaluating an AI copilot vendor for formulation support and customer correspondence. Before signing, the IT lead asked which Code chapters the vendor’s model provider had signed and requested the documentation. The provider had signed all three and supplied a current model card, closing the review without escalating to counsel. The check became a standard procurement step.
- Signatory status and chapter coverage recorded per vendor
- Documentation and copyright policy filed with the contract
- Review flag set to re-check status at contract renewal
- Escalation path to counsel defined for non-signatory vendors
Current developments and effects
The Code’s signatory landscape and link to binding enforcement are both still moving.
Uneven adoption across providers
Adoption split along model scale and jurisdiction, with some providers signing selectively.
- Several large US and EU-based providers signed all three chapters in 2025
- At least one provider signed only the Safety and Security chapter
- At least one major provider declined to sign at all
Enforcement timeline tightens around the Code
GPAI obligations have applied since 2 August 2025, and the AI Office gains full fining powers over GPAI providers on 2 August 2026. Signatory status now reads as a risk signal, not a decorative badge.
Interplay with harmonised standards
Unlike harmonised standards, still being drafted and not yet cited in the Official Journal, the Code of Practice is already operational, making it the more mature route for now.
Conclusion
The GPAI Code of Practice does not replace the EU AI Act, and signing it does not exempt a model from scrutiny. It gives the market a faster way to separate vendors that document their compliance posture from those that do not, as enforcement powers become real. Checking a vendor’s signatory status is a small procurement step that pays off once AI Office fining powers activate in August 2026. Review processes built around the Code now need only light upkeep later.
Frequently Asked Questions
What is the GPAI Code of Practice and who is it for?
A voluntary framework the Commission published in July 2025 for providers of general-purpose AI models, not for companies that use those models. Providers sign it to show compliance with transparency, copyright, and safety obligations.
Does a German Mittelstand company need to sign the GPAI Code of Practice itself?
No. The Code addresses GPAI model providers, not businesses accessing models through an API or product. A Mittelstand company checks whether its vendors signed; it does not sign itself.
How does the Code of Practice differ from the EU AI Act itself?
The AI Act is binding law creating the obligations. The Code is one voluntary way a provider shows it meets them; a provider can be fully compliant without signing, through another route.
What happens if a GPAI provider does not sign the Code?
Nothing automatic, but the provider must prove compliance another way and faces closer AI Office scrutiny, lacking the shortcut signatories use.
How can a company check whether its AI vendor has signed?
The Commission publishes a signatory list on its digital strategy site, and most signing providers state their status in their own trust documentation. Platforms built on third-party models, including Superkind, typically document which providers they use and what those providers signed.
What does checking Code of Practice adherence cost a company with under 250 employees?
Mainly staff time: a few hours per vendor to request documentation and record status in a checklist. The heavier documentation burden sits with the provider.