Isometric library with a figure reading a long scroll beside a balance scale and a glowing cube with a wax seal, representing AI model licence terms

Quick answer

Usually yes, you can legally modify an open-weight AI model, but only within the terms of its licence, and you remain responsible for how you use what it produces. Permissive licences such as Apache 2.0 allow modification with few conditions, while several big-company licences add acceptable use rules that carry over to modified versions. Before you edit or share a model, read its licence, keep a record of what you changed, and do not use the result for anything unlawful or harmful.

Key takeaways

  • The licence travels with the model. Editing it does not remove the original terms.
  • Licences differ a lot. “Open” does not always mean “do anything you like”.
  • You own the consequences of what you publish or send to customers, whatever model wrote it.
  • Sharing a modified model often brings extra obligations, such as passing on the licence.

This guide follows on from Uncensoring an AI Model: What Mark Kashef’s /unlock Skill Really Does, and the Risks, based on a video by Mark Kashef showing how tools like Heretic edit a downloadable model to change its behaviour. Before anyone in your business tries something similar, it helps to understand the legal side. This is general information for UK readers, not legal advice.

📘 Main article: Read what “uncensoring” a model means and the practical risks in Uncensoring an AI Model: What Mark Kashef’s /unlock Skill Really Does, and the Risks.

Can you legally modify an open-weight AI model?

In most cases, yes, if the licence allows it and your use is lawful. The model’s licence is a contract-like set of permissions. If you break it, you may lose the right to use the model and face claims from the developer. Separately, the law still applies to whatever you do with the output, such as data protection, consumer protection, defamation and online safety rules.

What is an open-weight model?

An open-weight model is one whose trained files (weights) you can download and run on your own computer, such as models from the Qwen, Llama, Gemma or Mistral families. That is different from “open source” in the strict sense, which would also include training data and code. Hosted tools like ChatGPT, Claude and Gemini in the browser are not open-weight; you can only use them under their terms of service.

What kinds of licences are there?

Licence typeTypical examplesWhat it usually means
PermissiveApache 2.0, MIT (used by many Qwen and Mistral models)Use, modify and share freely, keeping licence and notices
Community / customMeta Llama licence, Gemma terms of useModification allowed, but with acceptable use rules and sometimes conditions for very large companies
Research / non-commercialVarious research releasesNo commercial use, sometimes no redistribution

⚠️ Warning: Licences change between model versions. Always check the licence on the exact model page you download from, not a blog post (including this one).

What changes when you modify a model?

  • The original terms still apply. Community licences often say their acceptable use rules apply to derivatives too.
  • Sharing triggers obligations. If you publish an edited model, you usually must include the licence and say it was modified.
  • Naming matters. Some licences restrict using the developer’s brand name for your modified version.
  • Safety changes can breach terms. Removing safeguards to enable uses the acceptable use policy prohibits is likely to break the licence.

Who is responsible for what the model says?

You are. If an edited model writes something defamatory, misleading or harmful and you publish or send it, the responsibility sits with you or your business, not with the model or the tool you used to edit it. That is one reason we recommend keeping experimental models away from customers.

If you handle personal data with any AI model, UK GDPR still applies. A local model can help with privacy because data stays on your machine, but you still need a lawful basis and appropriate security.

What should you check before modifying a model?

  1. Read the licence on the official model page and save a copy.
  2. Check whether commercial use is allowed.
  3. Read any acceptable use policy and confirm your plan fits it.
  4. Note exactly what you changed and why.
  5. Decide whether the edited model will stay internal or be shared.
  6. Keep the original model so you can compare and roll back.
  7. Ask a solicitor if the model will be used in a product or with customers.

What does this mean for a small business?

For most small businesses, modifying a model is unnecessary. If a model refuses legitimate work, try the simpler fixes in our guide first, and set clear rules for your team with an AI acceptable use policy. If you do experiment, keep it internal, documented and away from customer-facing work until you are confident in both the quality and the legal position.

Frequently asked questions

Is it legal to uncensor an AI model?

Modifying an open-weight model is not automatically illegal, but its licence still applies and some licences forbid uses that removing safeguards would enable. You are also responsible for the output.

Can I sell a product built on a modified open-weight model?

Sometimes. It depends on the licence: permissive licences usually allow it; others restrict commercial use or add conditions. Check the exact licence and take legal advice.

Does running a model locally mean the licence does not apply?

No. The licence applies however and wherever you run the model.

What is the difference between open-weight and open source AI?

Open-weight means you can download the trained model files. Fully open source usually also means the code and training data are available under an open licence.

author avatar
Garry Knight
I'm Garry Knight, the person behind Prodify Digital. I write about email list building, email marketing, SEO, AI search and the tools that connect them. My aim is to make online marketing easier to understand, so creators and small business owners can make informed decisions about building an audience and keeping people engaged. Here you'll find straightforward guides and product reviews that explain what something does, where it fits and which limitations matter. The focus is on clear explanations and useful next steps—not hype, shortcuts or promises of easy earnings.

Leave a Reply