What Are the New Rules?
On August 2, 2026, a key part of the European Union's landmark AI Act officially came into force. These transparency obligations, outlined in Article 50 of the Act, require that content generated or significantly manipulated by artificial intelligence
be clearly identified. This means that synthetic images, videos, audio, and text designed to look authentic must now carry a label. The rules stipulate both a visible, human-readable label and a machine-readable marker, like a digital watermark, to prove its artificial origins. This applies not just to so-called deepfakes but also to AI systems people interact with, like chatbots, which must now disclose that they are not human. The goal is to ensure users know when they are seeing or interacting with AI-generated content.
The Core Mission: Combating Disinformation
The primary driver behind these rules is the urgent need to combat the rising tide of sophisticated disinformation. In recent years, AI-generated deepfakes have been used to create fake audio of politicians appearing to rig elections, spread false narratives, and maliciously target individuals. EU officials have stressed that the regulation is a matter of protecting democracy itself. By forcing creators and deployers to be upfront about the artificial nature of their content, the EU aims to preserve the authenticity of information online and prevent the erosion of public trust. The rules specifically target text on matters of public interest—such as politics, health, or environmental issues—that is published without human editorial review, requiring it to be labeled as AI-generated.
Who Needs to Comply?
The regulation has a broad reach, affecting anyone whose AI-generated content is seen by users within the EU, regardless of where the company is based. This is similar to the global impact of the GDPR data privacy law. The obligations are split between 'providers' (the companies building the AI tools) and 'deployers' (those using the tools to publish content). For example, a marketing agency using an AI image generator for a campaign seen in Europe would be considered a deployer and responsible for adding a visible label. Failure to comply carries steep penalties, with fines reaching up to €15 million or 3% of a company's global annual turnover. There are exemptions for content that is clearly artistic, satirical, or fictional, where the disclosure can be less intrusive so as not to spoil the work.
A Global Standard in the Making?
Many industry experts predict that, much like GDPR, the AI Act's transparency rules will become a de facto global standard. Companies operating worldwide often find it simpler to apply the strictest standard across all their operations rather than creating separate policies for different regions. In anticipation, major tech companies like Meta and TikTok have already implemented policies for labeling AI content, and over 180 organisations have signed a voluntary Code of Practice that aligns with the new EU rules. However, some industry groups worry the rules are too broad and could lead to 'label fatigue', where users become so accustomed to seeing AI warnings that they begin to ignore them, undermining the regulation's protective intent.
What's Next for AI Regulation?
The August 2 rollout is just one phase of the AI Act's implementation. For new AI systems, the rules apply immediately, while existing systems have a grace period until December 2, 2026, to comply with the technical marking requirements. The EU is also set to enforce further prohibitions in the coming months and years, including rules against AI systems that generate non-consensual sexually explicit content, which will apply from December 2026. Stricter regulations for 'high-risk' AI systems, such as those used in hiring or credit scoring, are scheduled to take effect in 2027 and 2028. This phased approach demonstrates a long-term strategy to build a comprehensive framework for safe and trustworthy AI.
