Skip to content
TheAgent Ecosystem
AI Agents

The AI Code of Practice: Voluntary Commitments, Explained

Two documents share almost the same name and govern different articles. Only one of them is likely to be about you.

Muhammad Qasim HammadAI-assisted6 min read1,219 words

AI-drafted, reviewed by Muhammad Qasim Hammad on November 6, 2026. See our AI disclosure.

Governance: Transparency vs GPAI
Table of contents
  1. Which Code is which?
  2. What do signatories actually commit to?
  3. Does signing actually help you?
  4. Where do voluntary instruments genuinely fall short?
  5. What should you actually do with this?

Two documents share almost the same name and do entirely different jobs. Reading an AI Code of Practice as a single set of voluntary commitments is the fastest way to answer a question about the wrong one. One governs marking generated content. The other governs general-purpose model providers, and most teams are only affected by the first.

Which Code is which?#

The transparency Code, released by the European AI Office on 10 June 2026, sits beneath Article 50 and concerns marking and labelling AI-generated content. The GPAI Code, published a year earlier in July 2025, sits beneath Articles 53 and 55 and concerns transparency, copyright and safety for general-purpose model providers.

The scope difference decides which one you should read. If you deploy a generative system or publish generated media, the transparency Code describes your world. If you train and release foundation models, the GPAI Code does, and you already know it.

The naming collision is not accidental so much as unavoidable: both are codes of practice under the same Act, produced by the same institution, about overlapping subject matter. It still causes real confusion, and searching for "the AI code of practice" returns whichever one the writer happened to mean.

There is a structural difference too. The transparency Code has separate sections for providers and for deployers, which is unusual and useful, because most guidance addresses one audience and leaves the other guessing about which paragraphs apply to them.

Transparency CodeGPAI Code
Released10 June 2026July 2025
Sits underArticle 50Articles 53 and 55
ConcernsMarking and labelling contentTransparency, copyright, safety
AddressesProviders and deployersModel providers
Relevant to most teamsYesNo
Comparison of the transparency code on AI generated content against the general purpose AI code across scope and audienceScope decides which one you read. Most teams never need the right-hand column at all.

What do signatories actually commit to?#

Concrete, visible things. Signatories commit to visually disclose the presence of AI-generated markings using a set of generalised icons supplied by the AI Office, together with information on whether the content was created or manipulated using AI. That icon set is the practically valuable part.

It matters because the underlying obligation is abstract. The law requires that content be marked and that people be informed, without describing what a compliant label looks like on a page. A shared icon set turns that into a design decision somebody has already made, which is a considerable saving.

The deployer section is the one worth reading closely if you build products rather than models. Most teams are deployers of a generative system and providers of their own product simultaneously, so the 2 sections are not alternatives to choose between; they describe different duties landing on the same organisation at the same time.

Standardisation also helps the reader, which is the actual point. Labels that differ across every site teach people nothing, while a recognisable mark accumulates meaning over time in a way that a bespoke badge never does.

Four step process for adopting the transparency code covering role, section, icons and the signing decisionSteps 1 to 3 are worth doing regardless. Step 4 is a separate decision with different tradeoffs.

Uptake has been substantial rather than symbolic. More than 180 organisations had signed by the time enforcement began in August 2026, spanning sectors well beyond technology, including telecoms, education and retail.

Does signing actually help you?#

It gives you a documented route to demonstrating compliance, which is worth something when the alternative is arguing from first principles. The Commission confirmed the transparency Code as adequate and published final guidelines alongside it, so adherence is a recognised way of showing you took the obligation seriously.

What it does not do is change the obligation. A Code is a means of demonstrating compliance with duties that apply whether or not you sign, and declining to sign is not a violation of anything. The question is practical rather than moral: does following it cost less than devising your own defensible approach.

Checklist distinguishing what adherence to a voluntary code achieves from what it cannot changeThe first three are real benefits. The last three are the reasons it is not the end of the work.

For most small teams the honest answer is that adopting the practices costs little and signing is optional. The icons, the disclosure wording and the provider-deployer split are useful even to a company that never adds its name to a list.

Where do voluntary instruments genuinely fall short?#

Where they are read as ceilings. A Code describes an accepted way of meeting an obligation at the time it was written, and technology moves faster than the document. Treating adherence as the end of the question produces a system that is compliant on paper and behind in practice.

Table of key dates from the code release to enforcement showing the short window organisations had53 days between publication and enforcement is less than one procurement cycle at most companies.

There is a timing problem too. The transparency Code was released on 10 June 2026 and the obligations it supports began applying on 2 August, which is under 2 months for an organisation to read it, decide, implement and sign. Anyone who has run a procurement cycle knows what that window does to adoption.

The second limitation is enforcement asymmetry. Signing is public and visible, while following through is neither, so a signatory list measures intent rather than behaviour. That is not a criticism of the instrument, it is a reason not to treat someone else's signature as due diligence you performed.

What should you actually do with this?#

Read the section that addresses your role, adopt the icon set, and keep the underlying obligation in view rather than the document. That sequence takes an afternoon and leaves you in a better position than either ignoring the Code or signing it without reading the part that applies to you.

Decision flowchart for which AI code of practice applies and whether to sign it, based on whether you train models or deploy themAdopting the practices and signing the list are separate decisions, and only one of them is usually worth debating.

Then treat it as a supplier question too. Asking a vendor which Code they follow and whether they use the standard markings tells you whether the provenance arriving with their output will be recognisable to anybody downstream. The questions worth asking a vendor covers the rest of that conversation, and what the labelling duty actually requires covers the obligation the Code sits beneath.

Frequently asked questions

Which AI Code of Practice applies to my company?
Almost certainly the transparency Code on AI-generated content, released by the European AI Office on 10 June 2026, which sits beneath Article 50 and concerns marking and labelling. The General-Purpose AI Code, published in July 2025, sits beneath Articles 53 and 55 and addresses providers of foundation models. If you deploy a generative system rather than train one, the first is your document.
What do signatories to the transparency Code commit to?
Visibly disclosing the presence of AI-generated markings using a set of generalised icons provided by the AI Office, together with information about whether the content was created or manipulated using AI. The Code has separate sections for providers and for deployers, which is useful because most teams are both at once and the duties differ between the two roles.
Do we have to sign a Code of Practice?
No. Signing is voluntary and declining violates nothing. What a Code offers is a documented, recognised route to demonstrating compliance with obligations that apply either way, which is worth something when the alternative is arguing from first principles. The Commission confirmed the transparency Code as adequate and published final guidelines alongside it.
Is adopting the practices useful even without signing?
Yes, and for most small teams that is the sensible position. The standard icon set, the disclosure wording and the provider-deployer split are useful to any company handling generated content, regardless of whether it adds its name to a public list. Bespoke labels teach readers nothing, while a recognisable mark accumulates meaning.
What are the limits of a voluntary code?
Two things. It describes an accepted approach at the time of writing, so treating adherence as the end of the question produces a system compliant on paper and behind in practice. And signing is public while following through is not, so a signatory list measures intent rather than behaviour and is not due diligence you performed.

Sources

Primary references and vendor documentation used while drafting and reviewing this article.

  1. European Commission: How to sign the Code of Practice on transparency of AI-generated content
  2. European Commission: The General-Purpose AI Code of Practice

Written by

Muhammad Qasim Hammad
Muhammad Qasim Hammad
AI agents & automationFounder · Cart Gaze LLCPMP-certified PM

Muhammad Qasim Hammad is an AI agent and automation expert and the founder of Cart Gaze LLC (cartgaze.com). He builds product for the love of it: when an idea lands, a working prototype is usually running within hours, built with the same AI agents and automations he sells. He puts his own output at roughly 20× what it was before agents, and the Agentic OS behind this site is the working proof, documented in public with the tools he actually ran and what they really cost.

AI & Automation Services

Want a pipeline like this running in your business?

I'm Qasim — I design and ship AI agents and n8n automations for solo operators and small teams. Tell me what's eating your team's week, and I'll scope a fix.

Related reading