"Limited-Risk" is the second regulatory tier of the EU AI Act. Unlike "High-Risk" systems which require conformity assessments and risk management, "Limited-Risk" systems are primarily regulated through transparency obligations to ensure that users are not manipulated or deceived.

This category applies to AI systems where the primary risk is a lack of clarity regarding the nature of the interaction. The core principle is that humans have the "right to know" when they are interacting with a machine or viewing artificially generated content.

Categories & Specific Obligations (Article 50)

The AI Act mandates specific disclosures for four main types of Limited-Risk systems:

1. AI Systems Intended to Interact with Natural Persons (e.g., Chatbots):
   
   • Obligation: Providers must ensure the system is designed to inform the user that they are interacting with an AI, unless it is obvious from the context.
   • Goal: To prevent users from mistakenly believing they are speaking to a human agent.
2. AI Systems Generating Synthetic Content (e.g., Generative AI):
   
   • Obligation: Providers must mark the outputs (audio, video, text, images) in a machine-readable format so that they can be detected as artificially generated or manipulated.
   • Goal: To combat disinformation and ensure content provenance.
3. Deepfakes (deployer obligation):
   
   • Obligation: Deployers who generate or manipulate image, audio, or video content that resembles existing persons, objects, places, or events ("deepfakes") must disclose that the content has been artificially generated or manipulated.
   • Exceptions: Exceptions exist for evidently artistic, creative, satirical, or fictional works, provided the disclosure does not hamper the display or enjoyment of the work.
4. Emotion Recognition & Biometric Categorization:
   
   • Obligation: Deployers of emotion recognition systems (that are not prohibited) or biometric categorization systems must explicitly inform the natural persons exposed to them.
   • Note: Emotion recognition in the workplace or education is generally prohibited (Unacceptable Risk), so this "Limited Risk" obligation applies only to other contexts, such as marketing or safety.

Strategic Impact: For most organizations, this category represents the baseline for consumer-facing AI. While the compliance burden is significantly lighter than for High-Risk systems (no extensive technical files or quality management systems required), the transparency requirements are strict. Failure to properly label a chatbot or deepfake can result in substantial fines, emphasizing that transparency is the prerequisite for trust in the European market.