Follow ZDNET: Add us as a preferred source on Google.
ZDNET’s key takeaways
- Grok 2.5’s license blocks true open-source use.
- Musk’s “open source” claim amounts to open-washing.
- Other AI projects offer real open access and freedom.
Companies love to exaggerate about open-sourcing AI. It plays well with people, naive developers get excited, and stock buyers invest more cash in their businesses. There’s only one little problem: It’s not true.
First, Mark Zuckerberg claimed Meta Llama was open source. Now, it’s Elon Musk’s turn, as he claims that his AI startup, xAI, is open-sourcing Grok 2.5, last year’s large language model (LLM).
Also: X’s Grok did surprisingly well in my AI coding tests
“The xAI Grok 2.5 model, which was our best model last year, is now open source. Grok 3 will be made open source in about 6 months,” said Musk on X. This release comes with the complete model weights. Grok 2 is available to download on Hugging Face.
Why is Musk doing this?
Unofficially, it’s to get more people excited and buying into Grok over its competitors. This is classic open-washing, where the name of the game is to claim something is open source without actually open-sourcing the code.
Also: Open-source skills can save your career when AI comes knocking
Officially, it’s part of xAI’s push for transparency and broader developer participation in its code. If you improve the code, xAI will be happy to use your changes. Of course, that’s true of any open-source project. However, I quote from the Grok license:
You may not use the Materials, derivatives, or outputs (including generated data) to train, create, or improve any foundational, large language, or general-purpose AI models, except for modifications or fine-tuning of Grok 2 permitted under and in accordance with the terms of this Agreement.
Yeah, so there is that.
That’s not so open, is it?
As one person put it on Y Combinator, those limitations mean:
- “Not open ‘source’ because the source isn’t available.”
- “Not ‘open’ weights because there are restrictions on use.”
- “This model is weights-available. There’s nothing open about it.”
Exactly so.
Also: AI is creeping into the Linux kernel – and official policy is needed ASAP
Leaving aside the Open Source Initiative (OSI) Open Source AI Definition (OSAID), which Grok doesn’t come close to meeting, the code also fails by the more broadly accepted open-source definitions.
Specifically, it fails on these grounds:
- Commercial use conditioned on guardrails: Commercial use is only allowed if the user complies with all provisions in xAI’s Acceptable Use Policy. This adds restrictions not present in OSI-approved licenses.
- Restrictions on use: The license explicitly prohibits using the materials “to train, create, or improve any foundational, large language, or general-purpose AI models” except for limited modifications or fine-tuning of Grok 2 itself. These field-of-use restrictions violate OSI’s criteria of no discrimination against fields of endeavor.
- Termination clause: The license is terminated if a licensee files certain kinds of litigation, which is more restrictive than OSI-compliant licenses.
What can you use Grok for in practice?
You can run, study, and modify Grok 2.5. xAI says this opens the door for independent experimentation, potential improvements, and transparency in how advanced AI systems are built. There are numerous other, more open AI projects, such as Mistral, Phi-2, BLOOM, and GPT-OSS, where you can learn hands on about how AI really works.
Also: OpenAI returns to its open-source roots with new open-weight AI models, and it’s a big deal
So, if you want to work on Grok, go ahead. Knock yourself out. Have fun.
Just don’t fool yourself into thinking you’re working with open-source code or open weights. You’re not.