EU AI Act Article 50 transparency rules now bind deepfake and AI text publishers

What Article 50 requires

Transparency obligations under Article 50 of the EU AI Act took effect on 2 August 2026, turning clear labelling of AI-generated content from a voluntary practice into a legal requirement for organisations publishing in the EU. According to a dev.to overview of the change, the rules cover two main categories: deepfakes, defined as AI-generated or AI-manipulated image, audio or video that resembles real people or events, and AI-generated or manipulated text published on matters of public interest, subject to exceptions.

As dev.to explains, Article 50 combines two related transparency goals. AI-generated or manipulated material must be identifiable as such through machine-readable marking. Where content qualifies as a deepfake, or where covered text concerns a matter of public interest, users must also receive a clear disclosure.

The European Commission has published a voluntary Code of Practice for AI-generated content to help providers and deployers implement the binding requirements. It has also released a free set of EU Icons — three icons plus variations — intended to help audiences recognise synthetic material. The key distinction, according to the dev.to article, is that the Code of Practice is optional while the underlying transparency duties are not. The icons are meant to accompany disclosures rather than replace the broader requirement for machine-readable identification.

Timing, exceptions and penalties

New content covered by Article 50 falls under the requirements from 2 August 2026. Content already placed on the market before that date benefits from a grace period, with labelling obligations fully applicable by December 2026. Artistic works may qualify for exceptions, with proportionate disclosure depending on the circumstances of the content.

The rules do not prohibit AI use. They establish a transparency baseline around content that could mislead people about what is real, who said something, or whether an event occurred. As dev.to points out, that distinction matters for marketing teams: an AI-created visual in a campaign may require a very different assessment from a realistic synthetic video of a person making a statement.

The AI Act provides for penalties of up to €15 million or 3% of worldwide turnover for companies, plus other penalties for EU institutions, according to the source.

Operational steps for publishers

Businesses often assemble content from a mix of internal tools, agencies, freelancers, stock assets and social platforms, which makes it easy to lose track of whether a final asset was generated, edited or materially manipulated by AI. Article 50 turns that lack of visibility into a practical risk whenever the result falls within the regulated categories.

Dev.to suggests a starting point of mapping the content path from creation to publication: identifying which tools generate or alter images, audio, video and text, and establishing who decides whether a disclosure is required. Recommended operational steps include:

• Building an inventory of AI content tools used by employees, contractors and agencies
• Recording the origin and editing history of assets that may be published externally
• Adding a review step before publication for realistic AI-generated or manipulated media and relevant public-interest text
• Using the EU icons alongside required disclosures where they fit the content and distribution channel
• Reviewing content placed on the market before 2 August 2026 ahead of the December deadline

The change reaches beyond design or legal review into how content is requested, approved, stored, repurposed and distributed. A team may, for instance, need to preserve the information required for machine-readable identification when exporting a video for multiple platforms. Agencies and external creators may also need clearer delivery requirements so that clients can make informed publishing decisions.

The more immediate problem is often operational rather than legal: an organisation that cannot identify which content was AI-generated or manipulated will struggle to apply labels consistently or update old material within the grace-period timeline. A documented process can reduce that uncertainty without turning every creative decision into a lengthy compliance exercise, according to dev.to.

Why it matters

This is one of the first hard compliance deadlines in the EU AI Act that touches everyday content workflows rather than model developers alone. Any team shipping generative AI into advertising, social media, video, audio or public communications in the EU now needs provenance tracking, disclosure review and a plan for legacy content — and the December 2026 deadline for pre-existing material leaves limited time to build that visibility. The voluntary Code of Practice and free EU Icons lower the implementation cost, but they do not substitute for assessing obligations under Article 50 itself, and the penalty ceiling of €15 million or 3% of global turnover gives the requirement real teeth.