Companies doing business in the European Union are now legally required to disclose when content is created or altered by artificial intelligence, under transparency rules in the bloc’s AI Act that became enforceable this month. The obligations, detailed in a report from Euronews, apply to synthetic images, video, audio and text and mark one of the most significant efforts yet to force disclosure of AI-made material at scale.
What the New Rules Require
The transparency provisions fall under Article 50 of the EU AI Act, the sweeping law that entered into force in August 2024 and has been rolling out in phases. As of this month, companies that build or deploy generative AI tools must ensure their outputs carry a machine-readable marker identifying them as artificially generated or manipulated, a technique often referred to as watermarking.
The rules extend beyond static labels. Deployers of systems that produce deepfakes — realistic but fabricated images, audio or video of real people, places or events — must clearly disclose that the content is artificial. Businesses publishing AI-generated text on matters of public interest, such as news-style articles, must also flag it as machine-made unless the material has undergone meaningful human editorial review.
Chatbots and other AI systems that interact directly with people are separately required to inform users they are talking to a machine, unless that fact would already be obvious from the context. Limited exceptions exist for law enforcement uses tied to detecting or prosecuting crime, and for clearly artistic, satirical or fictional works, though even those cases require some form of disclosure that does not disrupt the display of the work.
Who Has to Comply
The requirements apply broadly to any company offering AI-generated content or AI-powered interactive tools to users inside the EU, regardless of where the company is headquartered — a structure similar to how the bloc’s GDPR privacy law reaches firms based outside Europe. That means major AI developers such as OpenAI, Google and Meta, along with smaller software companies embedding generative tools into their products, fall under the mandate if they serve European users.
Some large platforms had already begun rolling out labeling systems ahead of the deadline. Google has used its SynthID watermarking tool across image and audio outputs, and several major AI labs have adopted metadata standards developed by the Coalition for Content Provenance and Authenticity, an industry group building shared tools to trace a file’s origin and edit history. The new EU rules effectively make similar practices mandatory rather than voluntary for anyone operating in the bloc.
Penalties for Getting It Wrong
Noncompliance carries real financial exposure. Violations of the AI Act’s transparency obligations can trigger fines of up to 15 million euros or 3% of a company’s total worldwide annual revenue, whichever figure is higher — a tier below the steepest penalties reserved for banned AI practices, which can reach 35 million euros or 7% of global turnover. Enforcement is expected to run through national market surveillance authorities in each EU member state, coordinated with the bloc’s central AI Office.
Regulators have signaled that early enforcement will likely focus on the most visible offenders — large platforms and widely used AI tools — rather than smaller businesses experimenting with generative features. Even so, compliance officers across industries from advertising to journalism to customer service are now reviewing their AI tools to confirm labeling is built in correctly.
Industry Pushback and Practical Limits
Not everyone is convinced the labeling system will work as intended. Critics point out that watermarks embedded in metadata can be stripped out simply by taking a screenshot or re-uploading a file to a different platform, and that enforcement across a fast-moving global internet is inherently difficult for any single regulatory bloc to police. The rise of what critics call “AI slop” — a flood of low-effort synthetic content overwhelming genuine material online — has already created headaches well beyond Europe’s borders; the trend recently complicated Apple’s bug bounty program, where a genuine $200,000 macOS vulnerability report got buried under a wave of AI-generated submissions.
Some AI companies have also raised concerns that mandatory labeling could put EU-facing products at a competitive disadvantage if rivals in less-regulated markets ship features faster without the same disclosure overhead. European officials have countered that the rules are meant to build public trust in AI tools over the long term, particularly as synthetic media becomes harder for the average person to distinguish from authentic content.
What Comes Next
The labeling mandate is one piece of a broader rollout of the EU AI Act, which has already imposed bans on certain high-risk AI practices and separate obligations on developers of general-purpose AI models. Additional provisions covering high-risk AI systems used in areas like hiring, credit scoring and law enforcement are still scheduled to phase in over the coming years.
For now, companies serving European users have little choice but to adapt. Legal and compliance teams are expected to spend the coming months auditing AI-generated content pipelines, updating chatbot disclosures and integrating watermarking technology to avoid running afoul of a law that, unlike many national AI proposals still in draft form, is already in force.
