AI Transparency Obligations: What You Must Disclose to Users

Transparency is the minimum bar

Even if an AI SaaS product is not high-risk, specific transparency duties may still apply. Article 50 covers defined cases including direct interaction with natural persons, certain generated or manipulated content, emotion recognition / biometric categorisation, deepfakes, and some public-interest text.

It is not a universal duty to label every internal use of AI. Map the feature to the relevant Article 50 paragraph, actor, exceptions, and transition rules first.

Current source: the Commission published Article 50 implementation guidelines on 20 July 2026. The obligations apply from 2 August 2026, with only a limited transition to 2 December 2026 for certain pre-existing systems under the Article 50(2) marking and detection duty.

1. AI disclosure notice

Where: On your website (footer or about page) and in your product UI.

What to say:

"This product uses artificial intelligence to [describe function]. AI outputs are reviewed by humans before [action] is taken. For questions about our AI usage, contact [email]."

Why it matters: Failing to disclose AI interaction is the most common compliance gap we find in scans.

2. Privacy policy AI section

Where: Privacy policy, add a dedicated "AI and Automated Decision-Making" section.

What to include:

• Description of what the AI does
• What data feeds the AI model
• Whether decisions are automated or human-reviewed
• User rights: ability to opt out, request human review, appeal decisions
• Data retention for AI logs

3. In-product AI labeling

Where: Anywhere the user interacts with AI output.

Examples:

• Chatbot: "AI-generated response" label
• Content tool: "AI-generated" badge
• Recommendation engine: "Recommended by AI" notice
• Analytics: "AI-predicted value" annotation

4. Model documentation

Where: Internal (not public), but ready for regulatory review.

What to document:

• Model name and version
• Provider (OpenAI, Anthropic, open-source, custom)
• Training data sources and cutoff date
• Known limitations and biases
• Accuracy metrics (if available)
• Date of last evaluation

5. User rights communication

Where: Product UI and documentation.

What to communicate:

• Right to know they are interacting with AI
• Right to opt out of AI processing (where feasible)
• Right to human review of AI decisions that affect them
• Right to appeal or contest AI decisions

Common transparency mistakes

• No AI disclosure at the relevant interaction or publication surface
• Generic disclosure ("We use AI") without specifics
• Buried disclosure in TOS pages nobody reads
• No in-product labels — users don't know they're talking to AI

Quick win: add AI disclosure today

You can add an AI disclosure notice to your website footer in 10 minutes. It's the single highest-impact action for AI Act readiness.

For direct interaction, a footer alone may not be enough: put the notice at the point of interaction. See the July 2026 Article 50 guidelines summary.