Table of Content
Companies building or using AI-generated content tools face a major transparency deadline today, as the European Union moves closer to enforcing Article 50 of the AI Act on August 2, 2026.
Organizations that want to appear on the first list of signatories to the Code of Practice on Transparency of AI-generated Content must submit their completed forms by July 27 at 18:00 CEST. Signing later will still be possible, but late signatories will miss the initial published list released before the new rules begin applying.
The deadline arrives just days after the European Commission finalized its implementation guidance for Article 50 and confirmed that the Code of Practice has been assessed as adequate by both the Commission and the AI Board. That gives signatories a clearer route for showing compliance, while companies that choose not to follow the Code may need to prove their own alternative measures are sufficient.
What Article 50 Covers
Article 50 is focused on transparency. It does not depend on whether an AI system is classified as high risk. Instead, it asks whether people are being exposed to AI systems or AI-generated content without proper notice.
The rules split duties between providers and deployers. Providers are the companies that build or supply AI systems. Deployers are the people or organizations that use those systems in public-facing services, content, or workflows.
Providers must make sure users know when they are interacting with AI, unless that is already obvious. They must also mark synthetic audio, image, video, and text in a machine-readable way so it can be detected as AI-generated.
Deployers face separate duties. They must inform people when emotion recognition or biometric categorization systems are used. They must also clearly disclose deepfakes and certain AI-generated text published to inform the public on matters of public interest.
Not Every AI Caption Needs a Label
One of the biggest misunderstandings around the new rules is the idea that every AI-written text must be labelled. That is not how Article 50 works.
For text, the labelling duty applies when AI-generated or manipulated text is published with the purpose of informing the public on matters of public interest. That can include topics such as elections, public health, public security, law enforcement, fundamental rights, consumer safety, economic developments, environmental protection, or other issues that may form part of public debate.
A café’s Instagram caption, a creator bio, a product hashtag set, or a basic promotional post will usually not fall into that category. But a brand post about public health claims, election-related information, financial risk, consumer safety, or similar public-interest subjects could.
There is also an important human review exemption. AI-generated public-interest text does not need the same label if it has gone through meaningful human review or editorial control. But that review must be substantive. A spell-check, grammar pass, or purely formal approval will not be enough. Someone with relevant judgment must examine the content and take responsibility for approving, changing, or rejecting it.
Creators and Freelancers May Be Covered
The rules also draw a line between personal and professional use. A natural person using AI for a purely personal activity may fall outside the AI Act’s scope. But if the activity is tied to regular economic benefit, freelancing, business, trade, or professional work, that person may count as a deployer.
That distinction matters for creators. A hobbyist posting for fun is different from a monetized influencer, agency worker, freelancer, or brand account using AI-generated media in public-facing campaigns.
For companies, employees and contractors generally do not become separate deployers when acting under instructions. The organization remains responsible for the deployment.
Deepfakes Need Visible Disclosure
Image and video creators face a stricter issue around deepfakes. If AI-generated or manipulated media falsely appears authentic and resembles real people, places, objects, or events, it must be clearly disclosed to viewers.
Machine-readable metadata alone is not enough. The audience needs a visible disclosure that can be understood without special technical tools.
There is a lighter rule for content that is clearly artistic, fictional, satirical, or creative. In those cases, disclosure must still exist, but it should not ruin the audience’s enjoyment of the work.
Deadlines and Penalties
Article 50 begins applying on August 2, 2026. Content generated before that date does not need to be labelled retroactively, although voluntary transparency is encouraged.
A limited grace period applies to some AI systems already on the market before August 2, but only for the provider-side machine-readable marking obligation. Those systems must comply by December 2, 2026.
Penalties can reach €15 million or 3% of worldwide annual turnover, with proportionality considered for smaller businesses.
The rules also arrive alongside a similar transparency push in California, where large generative AI systems will face watermarking and detection requirements from August 2. California’s rules focus mainly on image, video, and audio, while the EU framework also includes certain public-interest text.
For creators, publishers, AI writing tools, and social media teams, the practical takeaway is clear: not every AI-assisted post needs a label, but deepfakes, public-interest content, and professional AI use now require much more careful handling.