# How We Manage AI

Canonical page: https://fuser.studio/legal/ai

Fuser is a creative studio built on generative AI. This page explains how we manage the risks that come with it. It summarizes our internal AI Use and Risk Policy, which follows the NIST AI Risk Management Framework (AI RMF 1.0).

## The AI in Fuser

Fuser runs third-party generative models for image, video, audio, 3D, text and code. You choose the model for each step of your workflow. An optional AI agent can build and edit workflows on your canvas, search for references and run generations for you.

The providers that process your content are listed on our [Sub-processors](https://fuser.studio/legal/sub-processors) page.

## Your content and model training

- We don't train or fine-tune AI models on your content.
- We opt out of provider training wherever a provider allows it. Any provider whose terms allow training on your content is named on our [Sub-processors](https://fuser.studio/legal/sub-processors) page and in our [Terms](https://fuser.studio/legal/terms), and workspace admins can block its models.
- Your inputs and outputs belong to you, as our [Terms](https://fuser.studio/legal/terms) describe.

## How we govern AI

- **Risk register:** AI risks are recorded in our company risk register, owned by our Security Officer, and reviewed at least once a year and whenever an AI feature or provider changes materially. Examples are prompt injection, provider data handling, model availability and harmful output.
- **Vetting providers:** every model provider is assessed for security and data handling before it receives customer content, and reassessed each year.
- **Reviewing features:** before an AI feature ships, we document what it does, what data it sends to which model, and what its output can trigger.
- **Training:** everyone at Fuser completes privacy and responsible-AI training when they join and every year.

## Controls you have

- **Workspace switches:** workspace admins can turn off each kind of AI (image, video, audio, 3D, chat, code and apps) and block individual models. These settings are enforced on our servers.
- **The agent:** the AI agent is off by default.
- **Usage limits:** every AI request is metered in credits, and the agent runs within step and credit limits.
- **Logging:** every AI request is logged with the user, workspace, project, model and time.

## Safety in the product

- We treat model output as untrusted. Anything an AI action changes goes through the same permission checks as the same action taken by you.
- We test AI features before release, including against prompt injection where a model reads outside content or uses tools.
- Model providers apply their own safety filters. Our [Community Guidelines](https://fuser.studio/legal/community-guidelines) prohibit harmful content, and you can report content that breaks them.

## Questions and reports

To report harmful, inaccurate or unexpected AI output, or to ask about how we use AI, contact support@fuser.studio. For privacy questions, contact privacy@fuser.studio.