EU AI Act reshapes chatbot design
Guidance on the EU AI Act says chatbot obligations depend on what the bot actually does—simple informational helpers usually require transparency, while bots that influence hiring, finance or education can be classed as high-risk with stricter rules. That means the same chat interface can trigger very different compliance requirements depending on whether it informs, recommends, or makes decisions. (getmyai.ai, qservicesit.com)
The European Union’s chatbot rules now hinge less on the chat window and more on the job the software is doing. A bot that answers questions can face disclosure duties, while a bot that screens job applicants or scores borrowers can fall into the bloc’s high-risk regime. (digital-strategy.ec.europa.eu, ai-act-service-desk.ec.europa.eu, ai-act-service-desk.ec.europa.eu) The Artificial Intelligence Act, Regulation (EU) 2024/1689, entered into force on August 1, 2024, after publication in the Official Journal on July 12, 2024. The European Commission says the law uses a risk-based system with four tiers, from banned uses to high-risk systems and lighter transparency rules for some tools such as chatbots. (commission.europa.eu, eur-lex.europa.eu, digital-strategy.ec.europa.eu) For ordinary chatbots, Article 50 says providers must tell people they are interacting with an artificial intelligence system unless that is already obvious to a reasonably well-informed user. The same article also requires synthetic text, audio, image, or video outputs to be marked as artificially generated or manipulated in machine-readable form when technically feasible. (ai-act-service-desk.ec.europa.eu) The line hardens when a chatbot moves from answering to affecting outcomes in school, work, or lending. Annex III lists systems used for recruitment and candidate evaluation, student admission and assessment, and consumer creditworthiness or credit scoring among the high-risk categories. (ai-act-service-desk.ec.europa.eu) That means the same conversational interface can land in different legal buckets depending on what sits behind it. A campus helper that explains deadlines is treated differently from a bot that ranks applicants, and a bank assistant that explains rates is treated differently from a system that evaluates a borrower’s creditworthiness. (ai-act-service-desk.ec.europa.eu, ai-act-service-desk.ec.europa.eu, digital-strategy.ec.europa.eu) The compliance calendar is staggered. The Commission’s implementation timeline says prohibited practices and artificial intelligence literacy rules started applying on February 2, 2025, general-purpose artificial intelligence model rules started on August 2, 2025, and Article 50 transparency rules plus Annex III high-risk rules start applying on August 2, 2026. (ai-act-service-desk.ec.europa.eu, digital-strategy.ec.europa.eu) High-risk status brings a much heavier checklist than a disclosure banner. The Commission says high-risk systems must meet requirements on risk management, data and data governance, technical documentation, logging, human oversight, accuracy, robustness, and cybersecurity. (commission.europa.eu, eur-lex.europa.eu) The law also bans some uses outright, including emotion-recognition systems in workplaces and education institutions and social scoring by governments or companies. The Commission published guidelines on those prohibited practices on February 4, 2025, as national authorities prepared for enforcement. (digital-strategy.ec.europa.eu, digital-strategy.ec.europa.eu) For product teams, the design question is no longer just whether a feature “uses chat.” Under the European Union’s rules, the harder question is whether the bot is informing a person, steering a choice, or making a decision that changes access to a job, a classroom, or credit. (digital-strategy.ec.europa.eu, ai-act-service-desk.ec.europa.eu, ai-act-service-desk.ec.europa.eu)
