code of conduct
in English: code of practice / code of conduct
A written code of conduct voluntarily adopted by AI development companies to demonstrate that they work responsibly and in compliance with laws.
In plain terms
A code of conduct is like a joint commitment by a professional association: joining is not mandatory, but those who sign it publicly commit to complying. In the AI world, the text is typically drafted by independent experts based on feedback from developer companies, researchers, civil society organizations, and authorities. It describes, for example, how to document a model, how to respect copyright, and how to handle security risks.
It is useful because the law usually only states the goal ('the model must be transparent'), while the code of conduct also shows how to achieve this in practice. Those who follow it can more easily prove to the authority that everything is in order.
The Hungarian text of the EU AI Act distinguishes between two related genres. The "code of practice" helps fulfill mandatory rules, primarily for developers of general-purpose AI models, but one is also being prepared for labeling AI-generated content. The "code of conduct" is for those who would adopt strict requirements even if not mandatory for them. In the news, the two are often referred to equally as codes of conduct.
How it works
- A law or international agreement states general expectations and invites stakeholders to work out the details.
- The text is prepared in multiple rounds: developer companies, researchers, civil society organizations, and authorities contribute. In the EU, the work is organized by the European Commission's AI Office, and the text is written by independent experts.
- Companies sign the finished code voluntarily, thereby publicly committing to its contents.
- The signer documents how commitments are fulfilled, and the authority takes this into account as compliance evidence.
- Those who do not sign must still comply with the law, but they must prove compliance with their own method, which usually involves more work.
An everyday example
You read in the news that one major AI company has signed the EU's code of practice for general-purpose AI models, while another has rejected it. This means the first has committed to the jointly developed rules on transparency, copyright, and safety. The other wants to prove it complies with the law via a different path.
Who is affected
A common misconception
Limits and risks
- The signature is voluntary, so not every company joins, and some commit to only certain chapters.
- Signing the code of practice does not mean the product is safe or flawless; it only means the company has committed to following the rules.
- The companies involved themselves participate in writing the text, so critics argue the commitments are sometimes milder than necessary.
- The technology changes faster than the code, so the text must be updated from time to time.
Similar, but not the same
- legislation (statute) – The law is mandatory for everyone and its violation is directly punishable, while the code is voluntarily adopted and is only one possible path to legal compliance.
- harmonised standard – Standards are issued by official standards bodies with detailed technical specifications, while a code is a faster-to-produce, more general list of commitments.
- ethical guidelines – Ethical guidelines state general values without accountability, while a code of conduct contains specific, verifiable commitments.
Related concepts
AI regulation, self-regulation, puha jog
Code of Practice for General-Purpose AI Models, code of practice on labelling AI content
EU AI-rendelet, general-purpose AI model, AI Hivatal, harmonised standard, conformity assessment
A brief history
Self-regulatory codes are an old tool: for example, the EU applied one against disinformation on online platforms in 2018. In the AI field, the G7 countries adopted the Hiroshima Process International Code of Conduct in October 2023. The European Commission published a practical code for general-purpose AI models, developed by independent experts, in July 2025.
Where things stand
Most major developers signed the EU code for general-purpose AI models, but some rejected it or only committed to one chapter. The underlying obligations have been in force since August 2025, and the AI Office can enforce them with fines since August 2026. A separate practical code on labeling AI-generated content began development at the end of 2025.
In the news
- GPAI Code of Practice transparency chapter published as Commission readies guidelinesSep 19, 2026 · Jog
- Leading AI labs propose slowdown and shared safety rulesSep 14, 2026 · Agent Age
- Microsoft drafts code of conduct on human control of AI and plans to triple computing capacitySep 10, 2026 · Agent Age
- EU AI Act transparency and labelling rules take effectAug 3, 2026 · Jog
Also known as
code of conduct · code of practice · code of practice · code of conduct · GPAI Code of Practice · GPAI Code · self-regulatory code · AI code of conduct
Sources
This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.