On 10 June 2026, the European Commission published the Code of Practice on Transparency of AI-Generated Content. On 2 August 2026, Article 50 of the AI Act will start to apply to providers and deployers of certain artificial intelligence (AI) systems. Any team using AI in user-facing interactions, content generation or publishing processes now has a clear date on the calendar. But the relevant question is not only what needs to be complied with. It is why this matters beyond compliance.

What changes from 2 August

Article 50 of the AI Act establishes transparency obligations for providers and deployers of certain AI systems. The rule is not limited to high-risk systems: it also applies to use cases such as chatbots, synthetic content, emotion recognition, biometric categorisation, deepfakes or AI-generated texts published on matters of public interest.

The obligations cover four types of situations: when AI interacts directly with people, when it generates synthetic content, when it is used for emotion recognition or biometric categorisation, and when it produces deepfakes or text published on matters of general interest.

The Code of Practice, although voluntary, translates these obligations into practical marking and labelling measures to facilitate compliance. However, it does not replace the AI Act or the guidance that may be issued by the Commission. Companies that sign it before 22 July 2026 may be included in the initial list of signatories and rely on its measures to demonstrate compliance from the start of Article 50’s application.

Why regulation is arriving now, and not earlier

Generative AI moved from experimental tool to business infrastructure in less than three years. The speed of adoption exceeded many organisations’ ability to manage their reputational, operational and social risks. The European regulator did not react to the technology itself, but to the scale of its use. In many cases, people interacting with AI-generated or AI-modified content have no clear way of knowing it.

That information asymmetry is the central problem Article 50 seeks to address. The proliferation of deepfakes and synthetic content in public debate also accelerated political pressure to establish clear criteria before the potential harms became harder to reverse. The Code of Practice was developed with the participation of more than 180 ecosystem stakeholders, including providers, deployers, academia and civil society.

What this means for startups and corporates

For teams building products with generative AI, the most relevant change is not technical, but one of product design. Transparency towards the user must be embedded from the start, not added as a compliance layer at the end. Informing users that a system is AI should happen at the point of interaction, not be hidden in the terms and conditions.

For early-stage startups, documenting the flows in which AI is involved is not just an internal obligation. It is a competitive advantage against players that have not yet incorporated it, and it provides evidence in the event of future regulatory reviews.

For corporates, the challenge is to audit content production processes. The key criterion is determining when human supervision is sufficient to apply a potential labelling exemption, and when it is not.

The frictions the code does not solve by itself

Technical marking of synthetic content, including watermarking, metadata or cryptographic provenance methods, raises a real interoperability challenge. Systems from different providers still do not share common and sufficiently consolidated standards.

For certain generative systems already placed on the market before 2 August 2026, the calendar provides for a transitional period to adapt compliance until 2 December 2026. In any case, implementation must be assessed on a case-by-case basis, especially when systems are already in market or embedded in business processes.

In addition, the exemption for content subject to genuine editorial oversight introduces a relevant grey area. It only applies when there is clear and documented editorial responsibility; a superficial review is not enough.

For startups with limited resources, the risk of assuming that simply reviewing AI-generated text is sufficient is real. Voluntary adherence to the Code of Practice provides guidance, but the legal obligation remains Article 50.

Transparency as a condition for sustainable adoption

Regulation formalises something the market was already penalising informally: the loss of trust when users discover they were interacting with AI without knowing it. Article 50 does not invent a new expectation. It turns it into a more concrete obligation.

For teams building with AI, the most useful reading is that transparency is a design criterion, not a compliance cost. Organisations that integrate it from the beginning will build a trust base that is harder to replicate than any technical feature.

For startups and corporates, embedding transparency into product design can make the difference between launching an AI feature and building a truly scalable product.

At GCO Ventures, we continue to analyse how regulation, technology and new business models are redefining the conditions for building and scaling startups. Find more insights on our blog.