How we use AI

This is a translation provided for convenience. If it differs from the Polish version, the Polish version applies. Polski

Version of 2026-11-14.

What the assistant does

Verdiktum uses an automated assistant based on a large language model. It can:

• write a short briefing for a case and suggest what to check next,
• refresh that briefing as new evidence arrives,
• answer a question when someone mentions it in a discussion,
• draft a new case on the request of an administrator; such a draft is published only after a person reviews it.

The starter cases published at launch were prepared with AI from public sources. They are marked as starter cases and their entries carry the AI label.

What the assistant does not do

• It does not add evidence to community cases.
• It does not rate evidence and has no influence on the community meter.
• It does not decide verdicts and does not make moderation decisions.

Labelling

Everything written by the assistant carries a visible “AI” label. Its account is marked as automated.

Limits

Language models make mistakes. The assistant is instructed to rely only on what users have added to a case, but its text can still be inaccurate or incomplete. Treat it as a help in orienting yourself, not as a source. If you see an error, report it with the flag next to the content.

Data

To produce a text, we send the model provider the public content of the case: its title, description, the evidence and, for an answer, the comment that mentions the assistant. We do not send e-mail addresses or other account data. Provider: [AI PROVIDER].

Control

The Operator can switch the assistant off entirely or limit its functions at any time.