The Deadline Before the Deadline: What July 27 Means for Anyone Shipping AI-Generated Content

For the past two months, most of the conversation about the EU AI Act has pointed at a single date. On 2 August 2026, the European Commission gains its enforcement powers over providers of general-purpose AI models, with fines that can reach 15 million euro or 3 percent of worldwide annual turnover. That story is real, and it matters to a specific and relatively small group of model providers. It has also crowded out a second story that touches a far wider set of organisations, arrives on the same day, and carries its own earlier deadline.

That second story is Article 50, the transparency layer of the AI Act. It governs a simple-sounding idea with wide reach: people should know when they are dealing with a machine, and when what they are reading, seeing, or hearing was generated or altered by one. Anyone running a customer-facing chatbot, publishing AI-assisted images or video, or producing synthetic text on matters of public interest sits inside its scope. And there is a deadline before the deadline. Organisations that want to appear on the first public list of signatories to the Commission’s Code of Practice on transparency have to submit their form by 27 July 2026, at 18:00 CEST. That is this coming Monday.

The timing is worth stating plainly, because the Commission has been moving quickly. It published the final version of the Code of Practice on 10 June. It adopted the final Article 50 guidelines on 20 July, two days before this article. Both documents landed inside a three-week window ahead of the 2 August application date. The message from Brussels is that the transparency regime is ready, and the runway is short.

What Article 50 actually requires

Article 50 sets out four distinct obligations, and it helps to read them from the deployer’s side of the table rather than the lawmaker’s.

The first is interaction disclosure. If your organisation runs a chatbot or voice agent that a person could reasonably mistake for a human, you have to make clear that they are talking to an AI system. The second is machine-readable marking of generative output. Providers of generative AI have to mark synthetic images, audio, video, and text in a format that other systems can detect, so that a downstream platform or tool can recognise the content as AI-generated. The third is deepfake labelling. If you publish image, audio, or video content that has been artificially generated or manipulated to resemble real people, places, or events, you have to disclose that it is artificial. The fourth covers AI-generated text published to inform the public on matters of public interest, which carries its own disclosure duty, with narrow exceptions where a human retains editorial responsibility.

The reach is the point. Enforcement powers over GPAI providers concern the companies that build frontier models. Article 50 concerns everyone who deploys the output of those models in front of an audience. For a mid-sized European company, the GPAI enforcement date may be someone else’s problem. The transparency obligations are almost certainly its own.

The Omnibus moved the big deadlines, and left this one standing

There is a reasonable question sitting underneath all of this. The Digital Omnibus on AI, the EU’s simplification package, was signed on 8 July and is expected in the Official Journal before 30 July so that it can take effect on 2 August. It deferred a large slice of the AI Act’s high-risk obligations. Stand-alone high-risk systems now have until 2 December 2027. High-risk AI embedded in products already covered by EU safety law has until 2 August 2028. If the Commission was willing to push those dates back, why not the transparency rules as well?

It did not. The transparency layer stayed on its original schedule. Article 50 applies from 2 August 2026, unchanged. That contrast is the real headline for most organisations. The parts of the AI Act that require heavy conformity work got more time. The part that asks you to be honest about when content is synthetic did not. Reading the Omnibus as a general reprieve is the mistake to avoid this summer.

One more distinction belongs here, because industry coverage keeps blurring it. The GPAI transparency and documentation obligations have been in force since 2 August 2025. What changes on 2 August 2026 is that the Commission acquires the power to supervise and to fine. The obligations are a year old. The enforcement is new. Anyone writing or briefing on this should keep the two apart.

The Code of Practice, and whether to sign

The Code of Practice on Transparency of AI-Generated Content is the voluntary instrument that sits alongside the binding rules. It was developed through a multi-stakeholder process that ran from September 2025 and drew in more than 187 participants from industry, academia, civil society, and the member states, with independent experts leading the drafting. It focuses on the technical machinery of two obligations in particular: the marking and detection duties under Article 50(2), and the labelling duties under Article 50(4).

It is worth being precise about what signing does and does not do. The Code does not replace the AI Act, and it does not replace the Commission’s Article 50 guidelines. It offers signatories an EU-recognised, practical framework for demonstrating that their marking and labelling measures meet the standard. In a compliance assessment, being able to point to adherence to a common, Commission-backed method is a stronger position than improvising one alone.

So the decision to sign comes down to an honest trade-off. On one side, a signatory gains a ready-made technical framework, a presumption of good-faith alignment, and visibility on a public list published before the obligations bite. On the other, signing means committing to verifiable operational practices and accepting the scrutiny that comes with a public commitment. For an organisation that already produces or distributes synthetic content at any scale, the framework is likely worth more than the exposure. For one that touches AI-generated content only at the margins, watching the first cohort of signatories before joining a later round is a defensible call. Either way, the 27 July deadline concerns one thing only: whether your name is on the first published list, or a later one.

A read for the days that are left

None of this requires panic, and it does not require a compliance programme built overnight. It requires a short, honest inventory, and a small number of decisions taken deliberately rather than by default.

Four questions carry most of the weight. Which of our systems fall inside Article 50, meaning any chatbot, any generative output we publish, any synthetic media, any AI-written public-interest text? Who inside the organisation owns the decision on whether to sign the Code of Practice, and do they have what they need to decide before Monday? What do we actually need, technically, to mark and label generative output in a way the guidelines will accept? And what do our contracts with model and tool providers already say about who carries the marking obligation, so that we are not rebuilding something a supplier is already handling?

An organisation that can answer those four questions is in a strong position for 2 August, whether or not it signs anything by 27 July. An organisation that cannot has found its priority for the week.

The transparency layer of the AI Act was always the part designed to be visible to ordinary people rather than to compliance teams. It is fitting that it is also the part arriving on schedule, while the heavier machinery slides to 2027 and 2028. For most European organisations, the question is no longer what the AI Act will eventually demand. It is what becomes true on 2 August, and whether there is a name to put on a list four days before.

OneSynergy advises organisations in Turin, Italy and across Europe on AI governance, regulatory strategy, and the operational side of putting AI to work responsibly.

OneSynergy works with organisations in Turin, Italy and across Europe on EU AI Act readiness and AI governance: obligation mapping, risk classification as scoping, transparency, governance and technical documentation, alongside your legal counsel. See how we work on the AI Act.

For the state of the rules today rather than on the day this was written, see EU AI Act: deadlines and current status, a dated reference page we keep current.