EU AI Act, from obligation to evidence
Whether your AI is high-risk, falls under the transparency obligation or sits in a GPAI model chain: the question is the same. How do you show it works responsibly? Pick your route and see what applies to you.
High-risk domains (Annex III)
AI in essential services
High-risk AI in credit scoring, insurance pricing, benefits and emergency triage. Risk check, evidence pack and human oversight for banks, insurers and public bodies.
AI in biometrics
High-risk biometric AI: remote identification, categorisation and emotion recognition. Risk check, evidence pack and human oversight for security and tech vendors.
AI in critical infrastructure
High-risk AI in energy, water, traffic and digital infrastructure. Risk check, evidence pack and human oversight for grid operators and utilities.
AI in education and training
High-risk AI in admission, assessment, examination and proctoring. Risk check, evidence pack and human oversight for educational institutions and edtech.
AI in work and workforce management
High-risk AI in recruitment, selection and workforce management. Risk check, evidence pack and human oversight for employers, HR-tech vendors and recruitment agencies.
AI in law enforcement
High-risk AI in law enforcement: risk assessment, evidence reliability and investigation. Independent review, safeguards and documentation.
AI in migration and border control
High-risk AI in migration, asylum and border control. Independent review, fundamental-rights safeguards and documentation.
AI in justice and democracy
High-risk AI in justice, dispute resolution and democratic processes. Independent review, transparency and documentation.
Transparency obligations (Article 50)
No high-risk AI? You may still fall under these transparency obligations.
Transparency for AI chatbots
Article 50 transparency obligation for chatbots and AI assistants. Correct user notification, recorded and review-ready, even without high-risk AI.
Labelling of AI content
Article 50 marking duty for AI-generated content. Machine-readable labelling, set up correctly and review-ready, even without high-risk AI.
Notification for emotion recognition
Article 50 notification duty for emotion recognition and biometric categorisation. Correct information to those involved, recorded and review-ready.
Disclosure for deepfakes
Article 50 disclosure duty for deepfakes and AI-manipulated media. Correct disclosure, recorded and review-ready.
AI text for the public
Article 50 transparency for AI-generated text informing the public. Correct disclosure or editorial safeguard, recorded.
GPAI model obligations
Building or integrating general-purpose AI models? A separate track sits next to Annex III and Article 50.