Let's start with Wikipedia, because the irony is almost too perfect to pass up.
In mid-May, the Wikimedia Foundation fired Brooke Vibber — the lead developer of MediaWiki since 2003, the first full-time employee the Foundation ever hired, and their first Chief Technical Officer. Over twenty years of institutional knowledge, walked out the door. A week later, on May 21st, they disbanded the entire Community Tech team — five engineers and a manager. That team's entire job was to build things the volunteer editor community actually asked for through an official channel called the Community Wishlist. The one team at the Wikimedia Foundation whose product owner was, in effect, the volunteers who built the thing.
Most of the people fired were union organizers.
Within hours, editors were signing a solidarity petition and threatening an editorial strike. The Wikipedia Signpost confirmed the disbanding the same day. It's apparently the first time Wikipedia's volunteer editors have ever organized in solidarity with paid Foundation staff.
The Foundation doing all of this is sitting on $296 million in reserves and a freshly profitable AI revenue stream from Wikimedia Enterprise — the program that sells high-speed access to Wikipedia's content directly to Amazon, Microsoft, Meta, Perplexity, and Mistral.
This isn't a money problem. It's a priorities problem. And Wikipedia isn't the only one who has it.
Wikipedia's volunteer editors — the obsessive, detail-oriented, citation-fighting nerds who built 65 million articles across 300+ languages — are the reason Wikipedia's corpus was worth selling to AI companies in the first place. Every fact, every citation, every heated talk page argument that converged on something resembling accuracy: that was free labor creating the product now being monetized.
The thank you for that is firing the team that existed to give those volunteers a voice, right after cashing the checks from the companies that wanted what those volunteers built.
There's also a longer-term problem buried here. AI companies trained on Wikipedia because it was one of the most reliably human-verified, cited, and contested repositories of knowledge on the internet. Destroy the volunteer community that maintained that quality, replace them with AI-generated content, and you degrade the very thing you just sold. The data feeding AI gets worse. The AI gets worse. The data gets worse. Round and round.
Somewhere down that spiral, an AI will confidently tell someone the Battle of Gettysburg happened in Antarctica, and it will cite Wikipedia, which will have gotten it from an LLM, which trained on the Wikipedia that used to be accurate. Nobody currently making these decisions seems to be the one who'll have to clean that up.
Here's the part where Wikipedia stops being a quirky one-off and starts being a case study.
There's a term going around right now — CEO AI psychosis — and it describes something very specific. CEOs are uniquely vulnerable to it because they're far enough removed from the actual work that when they see an AI demo, they only see the happy path. The impressive part. The "look what it can do" part. They don't see the next ten or twenty steps where things get complicated, where it doesn't scale, where the edge cases pile up, where the humans who used to handle those things quietly aren't there anymore.
They see a demo. They get stars in their eyes. They order implementation. They move on.
Y Combinator CEO Garry Tan described it at SXSW as "cyber psychosis" — said he'd been sleeping four hours a night from the excitement, and claimed a third of the CEOs he knows have it. His assistant said he was joking. He wasn't joking.
Cory Doctorow put it more pointedly: the CEO delusion that they're worth thousands of times more than their workers makes them uniquely easy prey for AI salespeople pushing them deeper into that delusion. Because AI that can replace workers is AI that justifies the salary gap. The pitch sells itself.
The scale of what this is producing is already staggering. In just the first five months of 2026, the tech industry has nearly matched all of 2025's layoffs — 115,430 people fired from 152 companies, with AI cited as the reason by the bulk of them.
Wikipedia is a good hook because the irony is sharp and the timing is fresh. But the pattern predates this by a couple of years and is accelerating.
OpenAI is the most naked version. Founded as a nonprofit with a stated mission of ensuring AI benefits humanity — I'll pause while you finish laughing — it has spent the last several years systematically dismantling that structure in pursuit of a valuation that now reportedly sits at $852 billion and is targeting $1 trillion at IPO. The mission became the marketing. The nonprofit wrapper became the thing they were escaping rather than honoring. I've covered the paper trail on this extensively in the TheranasAI series — the gap between what the documents say and what the marketing says is considerable and worth reading. And the nonprofit conversion fight got so messy it ended up in federal court earlier this year, with Elon Musk suing Sam Altman over it — which I covered in Evil vs Evil. The short version: nobody in that courtroom had clean hands, and the mission was the last thing either of them was actually fighting about.
Mozilla is the more complicated and more sympathetic case. In December 2025, new Mozilla CEO Anthony Enzor-DeMeo announced plans to turn Firefox into a "modern AI browser" — and users did not take it quietly. The backlash was immediate and loud enough that Mozilla had to promise a killswitch, which did ship in Q1 2026. The feature set keeps expanding anyway. The thing is — I actually get the strategic logic here. Mozilla's Google search deal is existential. Something close to 90% of their revenue. If antitrust rulings kill that arrangement — which the DOJ case made more likely — they need a plan B and they needed to have been building it yesterday. The AI pivot isn't necessarily "we want to enshittify Firefox." It might genuinely be "we are terrified and scrambling."
But the execution problem is the same regardless of intent. Firefox's core user base isn't casual users who just want a browser. They're people who switched specifically because they care about privacy, user control, and not being the product. They have strong opinions, they know how to read a privacy policy, and they're very loud on exactly the platforms where this stuff gets amplified. Annoying that demographic in service of a revenue strategy built for a different demographic is a specific kind of unforced error.
The underlying lesson in all three cases is the same: AI money is large enough relative to what these organizations were built to handle financially that it warps priorities almost immediately. Survival pressure corrupts mission just as reliably as greed does. The road to enshittification is paved with "we had no choice" as much as deliberate bad intent.
Here's the part that I think gets underreported because it requires connecting a few dots that don't live in the same news cycle.
The AI companies — OpenAI, Anthropic, Google DeepMind, the rest — are carrying valuations that require continuous growth narratives to sustain. Enterprise adoption. Proof of ROI. Testimonials. Case studies. The thing you put on the slide deck that says "look how much our customers are saving."
Every CEO who watches a demo and fires a team is a data point in that deck. Every organization that cuts humans in favor of AI subscriptions is proof of product-market fit. Every "we replaced our community tech team with an LLM" announcement — whether that's the framing or not — feeds the story that makes the next funding round, the next enterprise deal, the IPO narrative possible.
OpenAI's IPO ambitions are well past rumor at this point — they're targeting September 2026, with confidential S-1 filings reportedly already in motion. Anthropic is reportedly racing them to it, targeting late 2026 as well. The pressure to show growth — real, measurable, enterprise-scale growth — doesn't disappear because the mission sounds good. If anything it intensifies, because the gap between the valuation and the actual revenue requires an increasingly aggressive story.
So when the Wikimedia Foundation board installs a new CEO and she starts moving fast on cost cuts and structural changes within months of arriving — with the union organizers conveniently among the first to go — it's worth asking who that performance is for. The volunteers? Or the enterprise partners and the broader AI industry narrative that nonprofits too are buying in?
This is the part that keeps getting skipped over in coverage of these decisions, and it's the most important part.
The humans being replaced weren't just doing tasks. They were doing judgment. They were the error correction layer that operated through caring, through institutional memory, through the social reality of being wrong in public and having someone call you on it.
Wikipedia's volunteer editors didn't just write articles. They argued. They fact-checked each other. They demanded citations. They noticed when something felt off and went looking for why. That adversarial collaborative process — messy, sometimes petty, occasionally maddening — is genuinely good at converging on accuracy over time. It has a feedback loop. It has stakes.
An LLM has confidence. Which is almost the opposite of what you want in an encyclopedia. It'll tell you something wrong with exactly the same authoritative tone it uses for things that are true, because it doesn't have a "this seems weird, let me double check" reflex. It learned from what it was given, weighted toward consensus, and reports accordingly. If the inputs were good, great. If they weren't — and increasingly they won't be — it has no way to know the difference.
And despite what the industry very much wants you to believe, we are nowhere near the kind of AI that reasons its way out of that. We don't have Data or C-3PO. We definitely don't have R2-D2 — the one who improvised, reasoned under uncertainty, made judgment calls with incomplete information because the mission required it. R2 was capable of that partly because he was never wiped. His decades of accumulated operational experience were his intelligence. Every new AI model is essentially a wipe and retrain. The institutional memory doesn't carry forward.
What we have is a very articulate and very confident pattern-matching system that works impressively within its training distribution and hallucinates a bridge to familiar territory when it hits something outside of it. The industry is actively profiting from the confusion between what it is and what people imagine it to be.
Meanwhile the humans who could tell the difference are being handed severance packages.
Wikipedia's editors built the training data. The Foundation sold access to that data to AI companies. The AI money gave the Foundation the confidence to restructure. The restructuring targeted the union organizers and the team serving the community. The community is threatening to strike. If they do — or if they just quietly disengage — the quality of new Wikipedia content degrades. The AI that trained on old Wikipedia trains the next model on whatever fills the gap. The gap fills with slop.
The AI companies need the growth story to justify the valuation. The valuations need the IPO. The IPO needs enterprise adoption. The enterprise adoption is fueled by CEOs who saw a demo and got stars in their eyes and decided that the humans were the expensive part of the problem. One of those humans used to make sure the Battle of Gettysburg happened in Pennsylvania.
It's a machine that runs on hype and needs constant fuel regardless of whether the underlying reality supports it. And the fuel it's burning through right now includes some of the last load-bearing infrastructure of reliable information on the internet.
I'm sure that'll be fine.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。