Under EU regulations, AI labeling will be mandatory for content that resembles authentic material.

      Companies involved in the creation or utilization of AI will soon be required to label content that appears authentic but isn’t, as the European Union's groundbreaking AI Act transitions from principles to enforcement. Starting on 2 August, service providers and implementers throughout the EU will face mandatory transparency regulations regarding deepfakes, synthetic media, and machine-generated text, a development that platforms have been preparing for since YouTube initiated automatic labeling of AI videos.

      The fundamental concept is straightforward. When content has the potential to be perceived as real, individuals have the right to be informed that it was produced by a machine, and the responsibility now lies with the companies that create it.

      The regulations divide the responsibilities into two categories. Providers, the companies that develop generative systems, are required to incorporate machine-readable identifiers to ensure that images, audio, video, and text can be recognized as AI-generated or altered. Deployers are responsible for the more visible aspect. Anyone using AI to generate a deepfake or publish machine-written text on public interest topics must clearly and promptly disclose this origin to their audience.

      Having good intentions does not exempt one from responsibility. The requirement to label a realistic deepfake applies regardless of whether the creator intended to deceive, closing a loophole that might allow excuses like “we were only joking.”

      Chatbots are also included in these regulations. When an individual begins to interact with an AI system, they must be informed at the moment of interaction, rather than in the fine print of a service agreement that often goes unread.

      There are specific exceptions. Content that is clearly unrealistic or fantastical is not subjected to these rules, and AI-generated text does not require labeling if it has been significantly reviewed by a human with genuine editorial responsibility.

      Creative works will face a more lenient approach. While artistic and satirical content must still display a disclosure, it should be done in a manner that does not detract from the enjoyment of the work, acknowledging the needs of filmmakers and artists.

      The penalties for non-compliance are significant. Violation of these rules may result in fines of up to €15 million or 3% of a company’s global annual revenue, whichever is greater, which is enough to capture the attention of businesses in Silicon Valley and Brussels alike.

      The implementation timeline is somewhat flexible. While the obligations officially begin on 2 August, a separate simplification initiative has been proposed that could extend the machine-marking deadline to December, as part of broader modifications to the AI Act.

      The industry has been actively advocating for adjustments. Advertisers, among others, have contended that AI-generated advertisements should be exempt from these transparency requirements, a lobbying effort that coincides with the law's implementation.

      The Commission has sought to provide clarity. It released draft guidelines in May and is in the process of finalizing a code of practice for marking and labeling AI content, which companies can adopt to demonstrate compliance.

      The impact of these regulations extends beyond Europe. As global platforms typically do not create different systems for individual markets, the EU's labeling requirements are expected to influence how AI content is labeled worldwide, much like its previous privacy legislation did.

      Enforcement will present challenges. Watermarks can be removed, metadata can be lost, and machine-generated text is notoriously hard to identify, meaning the effectiveness of the rules will rely on the robustness of the detection tools available.

      Deepfakes represent a critical challenge. The EU has taken steps to prohibit non-consensual intimate deepfakes, and the labeling system aims to address the broader spectrum of synthetic media before it entirely undermines trust.

      The implications go beyond mere compliance. Europe is betting that mandatory transparency can help maintain a collective understanding of what is real online, especially at a time when generative technologies make creating convincing fakes increasingly easy.

      Whether this approach will be successful remains uncertain. This law represents the most ambitious effort thus far to label the synthetic landscape, but its effectiveness will depend on detection technologies evolving in tandem with generation methods—a race that, so far, the deceivers are winning.