Legal
Acceptable use policy
Last updated 2026-07-10
Most of this policy is enforced by the product itself, before a prompt ever reaches a model. This page states the rules in plain language.
Hard blocked, everywhere
The following are refused in every geography, with no override:
- Political and election content, including candidates, campaigns, and voting persuasion.
- Sexual content involving minors, and any sexualized depiction of real people without consent.
- Content designed to deceive by impersonating a real person or organization.
- Illegal content under applicable law.
Consent gated
Depicting a recognizable real person requires recorded consent. Without it, generation is refused by default. With it, safety scanning still applies to the output.
Provenance must stay intact
You may not remove, crop out, or obscure provenance manifests, watermarks, or disclosure flags in order to present synthetic media as authentic.
Platform rules still apply
When you publish to a platform through Cadenza, that platform's own policies apply on top of ours. Publish confirmation is always explicit and per destination.
Enforcement
Blocked prompts are refused before generation and logged. Evasion attempts, including obfuscated phrasing, are detected and treated as violations. Repeated or serious violations end beta access.