Use of AI

Normaro uses artificial intelligence in one clearly defined place. This page explains where it does and where it does not.

Where AI is used

• Report summary. A language model (Claude by Anthropic) summarises the measurement results, ranks them by urgency and writes suggestions for fixing them. These sections are labelled as AI-generated in the report.
• AI Act assessments. The model assesses from the page content whether a website shows features of a high-risk application under Annex III. The assessment is labelled as such and cites the relevant passages.

Where AI is not used

• Rule-based tests decide whether a criterion is violated (axe-core and our own measurements of contrast, zoom, keyboard and metadata), not the language model.
• AI does not make decisions on prices, payments or accounts.
• No chatbot answers you on this website. People answer the emails you send us.

Roles under the AI Act

Anthropic provides the language model as its provider; we use it as a deployer (Article 3(4) of Regulation (EU) 2024/1689). The machine-readable marking of AI outputs under Article 50(2) is the obligation of the model's provider. Our labelling of the AI-generated sections in the report is voluntary, because the reports are not intended to inform the public (Article 50(4)). The people who operate the Service have the AI literacy required by Article 4.

What data the model receives

Measurement results and text excerpts from the checked pages. It does not receive your account data. Under its commercial terms, Anthropic does not use this data for training. The privacy policy has the details.

Limits

Language models can make mistakes. Every statement in the report cites its source so that you can verify it. If you believe an AI-generated section is wrong, write to us: a human reviews it and corrects the report if necessary. The AI does not make decisions about you (Article 22 GDPR).