EU Chatbots Must Identify Themselves as AI-Generated by August 2026
A new regulation in the European Union is set to change how chatbots and other AI systems interact with people. As of today, any system that talks directly to individuals within the EU must inform them they are dealing with a machine. This requirement applies not just to newly developed systems but also those already live, marking a significant shift for companies operating in this space.
The regulation, which comes into effect on August 2, 2026, is part of the AI Act and focuses specifically on Article 50. According to the European Commission’s guidance, providers of chatbots and other interactive systems must ensure users are aware they are interacting with an artificial intelligence system from the outset. This includes clear labeling and transparency measures.
Four key areas are covered by this regulation: providers, deployers, generative systems, and labelling requirements. Providers of systems that interact directly with people must design them to inform users about AI involvement. Generative systems, which produce synthetic content such as audio or images, must be marked in a machine-readable format so it can be detected as AI-generated.
Deployers also have responsibilities under this regulation. They are required to notify individuals when emotion recognition or biometric categorization systems are being used and clearly label deepfakes and AI-generated text published for public interest purposes. This includes content generated before the August 2026 deadline, although retroactive labelling is not mandatory.
One of the key aspects of this regulation is its application to existing systems already in production. It’s not a rule that only applies to new developments but rather one that requires companies to adapt their current systems to meet these new requirements. The Commission has provided guidance on how to implement these changes, including a voluntary Code of Practice on transparency of AI-generated content.
The regulation also makes it clear that providers established outside the EU are in scope if their system’s output is used within the EU. This means companies operating globally must ensure compliance with this regulation even if they’re based elsewhere. The Commission has specified that fines can reach up to €15 million or 3% of total worldwide turnover, with proportionality available for SMEs and small mid-caps.
There are two exceptions to this rule: where it’s obvious a person is dealing with AI, and when content has had human review or editorial control. The Commission emphasizes the importance of transparency in these cases, stating that superficial checks do not count as sufficient human oversight.
The regulation also applies to consumer products such as customer-service bots and AI writing tools. Dating apps have been using AI features for over a year, and subscription platforms often employ chatbots to answer paid subscriber messages. This places certain corners of the internet squarely within Article 50’s scope.
For companies operating in this space, implementing these changes will be crucial. The Commission has given providers four months to adapt their systems before the December deadline closes the gap for existing systems. Watermarking synthetic text robustly remains an unsolved problem compared to watermarking images.