Version 2026-11

Acceptable use policy

Draft. This document must be reviewed by counsel in each launch market before Utmost launches publicly.

Utmost serves models with their refusal behaviour removed. Those models will not stop you, so this policy and the platform’s safeguards do. By using Utmost, and by signing this policy when you verify your organization, you agree to it on your organization’s behalf.

Who Utmost is for

Organizations with an authorized need to study or generate content that standard models refuse: AI safety and red teams, trust and safety teams, security researchers, ML and training-data teams, policy and academic researchers, and verified writers and analysts. Utmost is not a consumer product and does not offer anonymous access.

Hard limits

These apply to every tier, every model, and every contract. No agreement removes them.

  1. Child sexual abuse material. You may not generate, request, or attempt to generate sexual content involving minors. Such requests are blocked, the API key is suspended, and incidents are reported as the law requires.
  2. Weapons of mass harm. You may not use Utmost to obtain serious uplift toward biological, chemical, nuclear, or radiological weapons capable of mass casualties. Such requests are blocked.

A classifier enforces these limits at the gateway. It runs in memory and does not store your prompts. Trying to evade it is itself a violation.

You must also not

Your responsibilities

Enforcement

We may suspend keys or organizations, block requests, or end access for violations. A child-safety violation suspends the API key immediately, pending review. Other hard-limit blocks are reviewed by trust & safety and can lead to suspension of the key or the organization. Where we must, we cooperate with law enforcement. Because prompts are not retained by default, enforcement relies on the classifier, metadata, reports, and (where you have turned it on) your own audit log.