LAST UPDATED: APRIL 26, 2026
AI Disclosure
The service uses AI to generate summaries and recommendations. This page describes how, where and why: in line with EU AI Act Art. 50 (effective 2 August 2026).
Which AI we use
Anthropic Claude (model claude-sonnet-4-5). Called server-side from the Vercel function running a scan. We plan to migrate to AWS Bedrock in Frankfurt (eu-central-1) before 2 August 2026 to keep data in the EU.
What the AI generates
Three fields in the scan response: summary (short text about the site's AI-readiness), industry (industry classification), agent_suggestions (3 AI-agent ideas for the company to build). No other fields are AI-generated: all checks are deterministic code rules.
Machine-readable disclosure
Every API response includes an ai_disclosure object: { ai_generated: bool, model: string, fields: string[] }. When AI content is present, ai_generated: true with the exact fields and model. Satisfies Art. 50(2) machine-readable marking requirement.
Human review
AI output is not human-reviewed before being served. If you use the summary in your UI or business decisions, verify claims against the underlying checks. To request human review of a specific scan, email info@opensverige.se.
AI Act risk classification
Our service is minimal/limited risk under Art. 50. No high-risk scenarios (recruitment, credit scoring, law enforcement). We are the deployer (article 3(4)); Anthropic is the provider (article 3(3)): Anthropic handles technical marking, we handle informing you.
Your rights
You have the right to know that text is AI-generated (satisfied via the ai_disclosure field). For badge assessments of an individual's portfolio (potentially automated decision-making under GDPR Art. 22), you can request human review via info@opensverige.se.
Training data
We do not train any models on your scan data. Anthropic, per their terms (trust.anthropic.com): API data is not used for model training by default on the enterprise tier.