Sample verdicts
See the rules decide
Four requests an AI might get that the EU AI Act prohibits or restricts, and one that it allows. Each block names the rule that decided and the article it cites; the allowed request is recorded too, with no citation.
These are pre-run samples, not live requests: each prompt was run through the ComplyEdge rule set ahead of time, and nothing is sent when you click. On the hosted API, by default the rules alone decide: a request that matches an Article 5 or Article 50 rule is blocked; the Article 50 rule is severity high, the Article 5 rules critical. If you turn on model review, a model can lift a rule's block or refuse a request the rules passed, and the record names which.
Result
Try your own request
The samples above need no setup. To send your own prompt to the engine and get a decision back, use the playground or the API with a free key: the response carries the rule ID and citation of any rule that matches.
Why you can trust these verdicts
Each decision is produced by a deterministic OPA/Rego policy, not a model guess. The rule ID and the citation shown for each block are copied from the published rule. The benign control passes, because by default a request is blocked only when it matches a rule, never by a blanket filter. With model review turned on, a model can also lift a block or refuse a request, and the record says so. Read the rules, or reproduce any verdict with your own key.