Project Liberty
How Wikipedia is building a decentralized “immune system”
Imagine the following scenario.
- You are responsible for maintaining the accuracy of seven million English-language articles on the internet.
- You produce 500 new articles every single day, each one edited, fact-checked, and cited.
- Your review process is slow and methodical, sometimes taking weeks.
- The number of unpaid volunteers doing this work is declining.
- And you’re facing a flood of AI-generated text that makes the review harder.
This is Wikipedia in 2026. And rather than accelerate with AI, earlier this year volunteer editors voted to ban AI-generated text across the English-language site.
It’s the opposite of what most institutions are doing. In this newsletter, we explore what that decision means for the community-powered online encyclopedia, how they are using AI, and how the decision could serve as a blueprint for other movement-building and resistance to Big Tech.
Wikipedia’s decision
Earlier this year, editors of Wikipedia’s English-language articles (which make up just under 10% of all Wikipedia articles) decided to ban AI-generated text from the platform due to its violation of the site’s core content policies.
There were two exceptions:
- “Editors are permitted to use LLMs to suggest basic copyedits to their own writing, and to incorporate some of them after human review, provided the LLM does not introduce content of its own. Caution is required because LLMs can go beyond what is asked of them and can change the meaning of the text such that it is not supported by the sources cited. Examples of basic copyedits include spelling, punctuation, and capitalization.”
- “Editors are permitted to use LLMs to translate articles from another language’s Wikipedia into the English Wikipedia, but must follow the guidance laid out at Wikipedia:LLM-assisted translation.”
Outside of the editing and writing process, Wikipedia still uses AI tools. Chris Albon, its Director of Machine Learning, described how AI would be used in areas where it excels to “remove technical barriers to allow humans at the core of Wikipedia to spend their valuable time on what they want to accomplish, and not on how to technically achieve it.”
This includes:
- Supporting Wikipedia’s moderators with AI-assisted workflows that automate tedious tasks in support of knowledge integrity.
- Giving Wikipedia’s editors time back by improving the discoverability of information on Wikipedia to leave more time for human deliberation, judgment, and consensus-building.
- Helping editors share local perspectives or context by automating the translation and adaptation of common topics.
- Scaling the onboarding of new Wikipedia volunteers with guided mentorship.
The decision, voted on by 42 volunteer editors of Wikipedia (40-2), represents a democratic, distributed approach to digital governance that is often missing from how tech platforms set internal policies and dictate approaches to AI.
What led to the decision
Within a year of ChatGPT’s release, Wikipedia editors began noticing signs and patterns suggesting that AI-generated content was appearing across its millions of pages.
For example, the phrase “rich cultural heritage” emerged again and again, as did phrases like “nestled in the heart of” and “diverse array.” Wikipedia, which has since published a page, Signs of AI writing, has found more telltale signs of AI writing throughout its many pages.
The editors also faced an onslaught of AI-generated submissions, replete with questionable citations and false information.
This is a problem because of Wikipedia’s commitment to knowledge and accuracy. AI tools can change the meaning of something. For example, according to the Signs of AI writing page, “LLM writing often puffs up the importance of the subject matter by adding statements about how arbitrary aspects of the topic represent or contribute to a broader topic.”
A decentralized “immune system”
Wikipedia’s response, which encompasses far more than just its recent policy banning AI content, amounts to what Marshall Miller, the product director for the Wikimedia Foundation (Wikipedia’s parent organization), calls “an immune system.”
“[Wikipedia volunteer editors] are vigilant to make sure that the content stays neutral and reliable. As the internet changes, as things like AI appear, that’s the immune system adapting to some kind of new challenge and figuring out how to process it,” he said.
The immune system involves several protocols and cycles of review:
- Speedy Deletion: The Wikipedia editorial review process can take multiple days or weeks, but editors have installed a Speedy Deletion process for content that clearly violates the site’s rules. This enables the community of Wikipedians to remove AI slop quickly.
- Paste check and Edit check: Wikipedia has built tools that help new editors ensure their content aligns with platform guidelines. When an editor pastes 50 characters or more of unrecognized text, Paste check will prompt them to confirm that they wrote the content themselves.
- WikiProject AI Cleanup: Editors launched a project called AI Cleanup to equip editors with tools to detect AI content. Over 270 editors have signed up to volunteer to clean AI from the site.
An example of human-powered movement building
The response by the volunteer community of Wikipedians offers an example for others seeking to align AI to human goals. There might be some instances when we should fight AI with more AI—cybersecurity could be one—but there are others when the immune system to AI hallucinations, disinformation, and addiction is not more AI, but more humans in decentralized networks and movements—banding together to build institutions and knowledge worth trusting. As Wikimedia celebrated its 25 anniversary earlier this year, it made the case that Wikipedia’s approach to verifiability, neutrality, and transparency—in a world of increasing AI slop—makes it more valuable now than ever and more resilient to the rapid pace of change.
Wikimedia and its Wikipedian volunteers are not alone. Cross-organizational partnerships are forming around shared standards. Neighborhood coalitions are pushing back on planned data centers. Communities are experimenting with decentralized digital governance. Each one is a piece of the same human movement to keep AI accountable to human flourishing.
If you’re part of an effort like this, or know one we should learn from, reply to this email. We’ll be sharing more later this year about how this movement is taking shape.
Other notable headlines
// 🤔 An article in Rest of World asked, can we really keep kids safe online? Future of Privacy Forum CEO Jules Polonetsky says protecting minors online requires more than just restrictions and parental controls. (Free).
// 📱 AI is empowering a generation of vibe coders to build exactly what they want. The personal software revolution is here, according to an article in The Verge. (Paywall).
// 🤖 Meet the sad wives of AI. An article in WIRED asked, Are you married to a man who’s obsessed with AI? (Paywall).
// 💼 The AI backlash could get very ugly. An article in The Atlantic imagines what could happen if jobs actually start disappearing. (Paywall).
// 🙌 People are increasingly turning to chatbots for moral guidance. An article in Project Syndicate argued that AI developers must work together with faith communities to ensure that their systems embody the shared values that have long shaped human societies. (Paywall).
Partner news
// Confronting the ethical frontier of agentic AI
June 1 | Stanford, CA
Stanford’s McCoy Family Center for Ethics in Society is hosting a full-day Conference on Ethics and Agentic AI on June 1 at Stanford, bringing together philosophers, computer scientists, and tech leaders to examine the moral and social implications of AI agents. Register here.
// Reimagining media: Inside The Signal’s vision for trust-driven journalism
The Sustainable Media Center‘s Emma Lembke sat down with Hywel Mills of The Signal for a Substack Live conversation on the failures of the engagement economy and the case for media built on trust, rigor, and curiosity. Listen or watch here.
// 🍽 The tech bros are going to etiquette school, according to an article in The Wall Street Journal. Founders who built their names on coding and hard-charging leadership are learning that in the AI era, soft skills matter more than ever. (Paywall).
