This disclosure explains how Standard Intelligence uses AI in its platform. It covers the Navigator, which classifies an AI system against the EU AI Act, and Guided Discovery, which records evidence against each requirement.
Rules make the decisions
Rules compute every determination. They work from our regulatory corpus, the answers you confirm and the facts of your system. The AI may propose facts from your system description. A proposed fact is used only when you confirm it, or when your description states it and the AI cites the exact words. The AI does not decide how your system is classified, and the classification still works with the AI switched off.
Each classification is sealed as a record that cannot be changed. The record shows which model and prompt version wrote its explanations, and which version of the corpus it used.
What AI writes
AI writes only two kinds of text:
- the explanations beside citations; and
- the guidance that checks your Guided Discovery answers.
Everything else, including every determination, comes from the rules, the corpus and what you tell us.
The Navigator does not interact directly with people. The AI drafts text that appears in the product.
How we mark AI-written text
Before you first use the platform, it tells you which of its text AI writes. Every AI-written text then carries a machine-readable marker, in the product, through our API and in any export.
The models we use
The AI is Claude, from Anthropic. It runs through Amazon Bedrock in an AWS region in the EU, which keeps none of the data we send it. To find the relevant law, we also send your system description to Mistral, in the EU, which keeps it for no more than 30 days to monitor abuse.
We fix the exact model version in our code. We do not use your data, or any customer's data, to train or improve any model.
Your part
The platform is not legal advice. Review its outputs before you rely on them. Each obligation shows how far it has been reviewed: suggested by AI, reviewed by us, or confirmed by counsel.
If you publish AI-written text from the platform to inform the public, the EU AI Act may require you to say that it is AI-generated.
When something looks wrong
If an output looks wrong, tell us at support@standardintelligence.com. We treat a wrong output as a defect, fix it and check the fix.
Changes
We tell you in advance of a material change to how your analysis is produced. Urgent security or legal fixes are the only exception.
Contact
If you have any questions about how we use AI, email support@standardintelligence.com. Our privacy policy explains how we handle personal data, and our terms of service set out what we and you each do.
Standard Intelligence Ltd. Registered in England and Wales, company number 16737890. Registered office: 128 City Road, London EC1V 2NX.