← Back to GAT

Development-stage information · 20 November 2026

AI use & responsibility

How we use AI while developing GAT, and the requirements we are setting for the customer service.

Where the project stands. GAT is in development. This information site displays prepared renders; it does not accept prompts or uploads, generate assets or take payments. Requirements for the future service below describe our intended approach, rather than completed service features.

Where AI is used

We use AI assistance to help write software, interpret asset descriptions, prepare construction instructions and review some results. Blender builds and renders the objects from code, with material settings controlling their appearance.

AI can influence the shape and content of an asset through the instructions it writes. We describe that involvement openly. A Blender render does not, by itself, establish the permissions behind an AI model's training data or guarantee rights in an output.

The planned customer service will call a language-model API and run Blender on hosted infrastructure. Supported presets, saved construction instructions and ordinary technical checks should be reusable without asking a model to repeat the whole task.

Keep control outside the model

Development work is reviewed through rendered examples and technical checks. A successful script is one kind of evidence; a visual review and a tested game-engine import are separate checks. We only describe engine support after the relevant tests.

Before accepting customer jobs, the service needs:

  • Limits on allowed operations, job size, execution time and resource use, enforced by application code.
  • Isolated Blender jobs, restricted file access and separation between customer work and service credentials.
  • Technical checks and a preview that lets the customer inspect the result before export.
  • Versioned construction instructions, error reporting and a way to investigate or stop a failing workflow.

We do not treat a model's own assurance that its output is safe or correct as a substitute for these controls.

Creators, sources and permissions

We keep source and licence records for the components, textures and other material we include in GAT. The material demonstrations include procedurally created textures with recorded build information.

Our provider-selection work considers published training-data information, rights policies, contractual terms and available opt-out mechanisms. We cannot independently verify every source in a foundation model's training set, and we will not describe a model as universally consent-based without evidence.

For the customer service, customers will need permission to use any reference material they submit. We intend to provide a clear route for rights complaints, investigate specific reports and correct or withdraw affected material where appropriate.

AI assistance does not replace artistic judgment. Editable results, visible controls and human review are central to the tool we are building. Once GAT earns revenue, we will define how the project supports artists and workers, including its funding approach and how we report what it achieves. Until then, we do not claim that a funded support programme exists or promise a specific percentage.

Customer data and provider choices

This site does not collect asset prompts or uploads. Before the service opens, we will publish which providers receive customer data, what they receive, why it is needed and how retention and deletion work.

Our intended approach is to send only the information needed for a job, keep customer work separate from development data, and require a separate, explicit opt-in before reusing customer content to train or improve models. Provider-side handling also depends on the selected API and its configuration; those details must be verified before launch.

See the privacy notice for the information site.

Measure the complete job

AI requests, Blender processing, storage and delivery all consume resources. Our design aims to avoid unnecessary model calls by reusing reviewed operations and saved construction instructions. The amount saved needs to be measured.

We plan to assess energy use and emissions per completed asset, including failed attempts and retries. Estimates need a stated boundary, a method and uncertainty. Data-centre electricity sources and cooling practices matter, but a cooling claim alone does not establish the footprint of a particular job.

We do not yet have a measured per-asset footprint and do not claim that GAT is carbon neutral or environmentally superior to another service. We will publish useful measurements as the hosted workflow becomes available to test.

Responsibility includes responding

prototopiq network Inc. is responsible for how GAT is designed, described and operated. Our working research draws on the NIST AI Risk Management Framework and Canadian privacy guidance for generative AI. These are references for our work; we are not claiming independent certification.

We will review this policy as GAT, our provider choices, available evidence and relevant guidance evolve, and date material updates. If an example, rights issue or AI-related concern needs attention, email nirabhra12709@gmail.com or contact the prototopiq company page. Include the page or output concerned and enough detail for us to investigate.