Last reviewed: April 2026
In this age it feels important to make a formal statement about how Cutout Studios plans to use modern AI tools in a manner it deems ethical.
The overall goal for this policy is for it to both mitigate harm and affirm humanhood, while still enabling our studio to remain competitive.
At time of writing, there’s definitely a trade-off between “cruelty free” LLM usage and capability. Here’s our waterfall:
An “open-license” model is one that has been trained entirely on consent-given data. Leading open-license models are the ideal but so far have proven limited in their capabilities.
We will continue to monitor and eventually invest in improving the open-license ecosystem - while we aren’t experienced in training models, we are experienced in using code to wrangle LLM outputs.
We run models locally as much as possible to reduce environmental impact and keep data from leaking, and will continue to improve that local infrastructure where we can.
For the moment, the models best fitting our current capabilities happen to live in the Qwen family. No major model has provably clean training data, but Alibaba has at least published a Training Data & Governance Disclosure for the Qwen series.
Uncommonly there may be a qualifying development task for which a local model proves insufficient (typically due to the fact that we lack the hardware necessary to reliably run the right tool). Sparingly we prefer Claude here because, of the top providers, they expose privacy controls and are the most transparent:
In the main, we use LLMs to do things that simply were not possible before - these are what we call “Qualifying Development Tasks” (QDTs). The aim is to augment and focus our current capabilities, not replace them entirely.
Note that this list is not final and may evolve as the ecosystem does.
“Toil” is “stuff you don’t wanna do, but you gotta do”. These kinds of tasks often fall by the wayside or are skimmed over due to fatigue or disinterest. Examples:
In writing, there is proper grammar, formatting and flow that often takes multiple iterations to iron out.
In software, documentation and automated testing often become insufficient due to constant changes in the product. It’s too much to keep up and so these things are often dropped or become stale quickly.
Management at Cutout Studios doesn’t believe in arbitrary deadlines, but there may arise naturally occurring dates (an upcoming convention, major presentation, etc.) that prove existential to the company.