# Acceptable use

> What you cannot do with Deference, and what happens if you do.

Effective October 9, 2026 · Version 1.0

## At a glance

- **Enforcement:** We disable API keys used to harm people, attack the service or collect free credit twice.
- **Banned content:** No child sexual abuse material, weapons of mass harm, malware, fraud or harassment.
- **API keys:** Keep API keys inside your team. Reselling access needs our written agreement.
- **Limits:** Stay within rate, spending and key limits.
- **Providers:** Each model provider's usage policy applies to your requests as well.

## 1. What you cannot create

Do not use Deference to create, request or spread:

- sexual content involving minors, or anything that exploits or endangers children;
- help building biological, chemical, nuclear or radiological weapons, or other weapons of mass harm;
- content that promotes terrorism or violent extremism, or incites violence;
- malware, or help attacking systems you have no permission to test;
- fraud, scams, phishing, or impersonation of a person or organization;
- harassment, threats, or exposure of someone's private information;
- anything illegal where you or the people affected live.

Follow export control and sanctions law, including where your outputs go.

## 2. People decide about people

Do not let a model alone make decisions with legal or similar effects on people: credit, jobs, housing, insurance, education, medical care or legal matters. Keep a person in the loop who can review and change the result.

## 3. Rules for AI use

- Do not try to extract a provider's system prompts, model weights or training data.
- Do not work around a provider's safety systems.
- Do not use outputs to train models where the provider's terms forbid it.
- Label AI-generated content where the law requires it.

## 4. Keys and accounts

- Keep API keys private and inside your team. Never publish them in code, apps or screenshots.
- Do not resell access, or offer Deference as your own API, without our written agreement.
- Do not open extra accounts to collect free credit or faucet tokens twice, or to get around limits.

## 5. The service itself

- Do not probe, scan or load-test Deference without our written permission.
- Do not work around rate limits, spending limits or key limits.
- Use the test faucet for testing only.

## 6. Providers have rules too

Each model provider's usage policy applies to your requests on top of these rules, and a provider can refuse a request that breaks it.

## 7. Reporting

Report abuse or a security problem to [support@deference.so](mailto:support@deference.so). Include the request ID if you have one: it is in Activity and in the `x-request-id` response header.

## 8. Enforcement

When we believe these rules were broken, we can disable API keys, block the account from sending requests, or report the activity to the authorities. We tell you why unless the law forbids it. To appeal, write to [support@deference.so](mailto:support@deference.so).

## Version history

- October 9, 2026 · 1.0 · First published version.

Questions: support@deference.so