Transparency guide
Limited-Risk AI Transparency Requirements
Many commercial AI features are not high-risk, but still need clear user-facing transparency. For SaaS teams, the practical question is simple: does the user need to know they are interacting with AI or receiving AI-generated content?
Disclaimer: This page is educational and product-planning oriented. It is not legal advice. Use it as a checklist starter, then confirm obligations for your specific deployment.
When a transparency notice is usually relevant
| Feature type | Practical disclosure focus |
|---|---|
| Chatbots and AI assistants | Tell users they are interacting with an AI system unless the context already makes that obvious. |
| Generated text, images, audio, or video | Make synthetic or manipulated content clear when users may reasonably think it is human-made or real. |
| Deepfake-style media | Disclose that the content has been artificially generated or manipulated. |
| Emotion recognition | Inform exposed people when the system detects or infers emotions, subject to the use-case limits elsewhere in the Act. |
| Biometric categorization | Inform exposed people when the system categorizes them using biometric data, and screen for sensitive-category risks. |
Good notice content
- Use plain language near the AI interaction, not only in a privacy policy.
- Explain what the AI feature does and what it should not be used for.
- Tell users when outputs may be incomplete, inaccurate, or unsuitable for important decisions.
- Offer a human contact, review path, or correction channel when the output affects a user workflow.
- Keep a dated copy of the notice text used in each product surface.
Launch checklist for SaaS teams
- Map every AI surface: chat, recommendations, summaries, scoring, media generation, and internal admin views.
- Decide whether the AI role is obvious to an ordinary user.
- Place the notice where the user sees the AI feature, not after the task is complete.
- Check whether the feature also triggers high-risk, biometric, employment, education, or prohibited-practice review.
- Review the notice after model, data, or workflow changes.
Official sources
Last reviewed: July 3, 2026.