Wikipedia helped train your favorite AI, now the Wiki foundation wants a cut

[Alfonso Maruccia]

In context: In their never-ending quest to build profitable AI services, Big Tech and startups are constantly scanning the internet for fresh, human-made content. Wikipedia, the free encyclopedia at the center of this unprecedented infrastructure abuse, is now asking AI companies to pay for its content.

The Wikimedia Foundation has underlined the irreplaceable value of human-created knowledge in the era of chatbots and generative AI services. The organization overseeing Wikipedia and other major "Wiki" projects is also outlining its ideas about how AI companies should earn the right to access the encyclopedia's content – now and in the years to come.

Many studies have explained that generative AI cannot possibly exist in a vacuum. Developing and improving new large language models (LLM) requires textual content created by actual humans, as training LLMs on LLM-generated content will quickly make the entire AI infrastructure collapse because of the feedback loop effect.

As one of the largest projects centered around human-written content, Wikipedia has been one of the most important sources for AI development from the start. Modern chatbots would not exist without Wikipedia, Wikimedia said, but AI companies are not being "responsible" in how they use the encyclopedia's content.

Earlier this year, the foundation revealed that AI bots have become a significant issue for Wikipedia's reliability. The service has experienced a 50 percent increase in bandwidth used for multimedia downloads since January 2024, with clear signs of AI scraping detected since December 2024. Traffic has now doubled, and Wikimedia blames bots for at least 65 percent of the most resource-intensive network loads.

Therefore, the foundation is now asking AI companies to start acting responsibly toward Wikipedia. They can perform two "straightforward" actions to achieve this goal: attribution and financial support.

Generative AI services should give credit to Wikipedia's human contributions when using the encyclopedia's content, so that the service can keep attracting new volunteers in the future. Furthermore, AI developers should go through the Wikimedia Enterprise services to access the platform's datasets with its paid API.

The new API should provide AI companies with all the quality content they need, with no excessive impact on Wikipedia's server infrastructure. Thanks to proper attribution and financial compensation, the platform will continue to be a sustainable venture in the future, Wikimedia said.

Wikipedia is written and managed by volunteer communities, while the Wikimedia Foundation provides technology and legal support to the encyclopedia's infrastructure. The free online encyclopedia will be 25 years old in January 2026, with Wikimedia planning to use the money provided by AI companies to fund the next 25 years and beyond.

Permalink to story:

Wikipedia helped train your favorite AI, now the Wiki foundation wants a cut

 

[PXAbstraction]

100%. These companies are profiting off the work of volunteers. At the very least, those companies should help support the organizations that provide this information.

 

[JMGuy]

100% AI companies owe for Wikipedia’s content.

For years, they’ve freely scraped and turned that work into billion-dollar-valuation AI products with zero compensation or attribution. That’s exploitation, plain and simple.

Wikipedia’s servers strain under massive bot traffic from AI models that wouldn’t exist without the human knowledge hosted there.
Asking for payment and proper credit is a minimum. Free knowledge shouldn’t mean corporations get to use it for free profit while putting a non-profit out of business.

If we want Wikipedia to survive for the next 25 years, AI companies need to step up, pay for what they use, and respect the people who’ve made their tech possible to begin with.

 

[Neatfeatguy]

That doesn't surprise me based on the information that AI throws at people, what a joke. AI isn't AI, but what it is is only as good as the information it uses.

Use biased info, you biased answers.
Use wrong info, get wrong answers.

Using wikipedia is a way to get a lot of both. Sounds like these companies are willing to build their "AI" with poor information just to say that they have the best "AI".

 

[umeng2002]

Favorite AI? Quite presumptuous.

 

[NeonTrotsky]

When do volunteers who edited wikipedia for years with thousands of edits and article creations, get paid? Isn't this what the editors signed up for?

 

[PXAbstraction]

When do volunteers who edited wikipedia for years with thousands of edits and article creations, get paid? Isn't this what the editors signed up for?

They know it's a volunteer role and do that happily. AI companies take the information, costing the non-profit foundation millions in server costs and don't even give attribution, much less a donation for the massive profit they generated off the work of those volunteers.

 

[dmatheson]

Ai companies are stealing right and left. Artists are getting hammered and I've been writing my congressman over the last 8 years again and again and again. I went to a high-powered VC talk, and they agreed that they don't even look at a proposal unless it describes how it is going to use AI. I asked, "How are you handling 'fair use'?" They looked like freaking deer in headlights. After a big pause, they asked, "Are you a lawyer?" <curse word goes here>