So Microsoft’s own people have been sitting in meetings, looking at spreadsheets, and basically watching their AI models eat the internet alive. And they knew it. They called it a “doom loop” – which, honestly, is the most accurate corporate jargon I’ve heard in years.

They Saw It Coming and Built It Anyway

Internal documents that surfaced in The New York Times’ lawsuit against Microsoft and OpenAI show that executives weren’t just aware of the problem. They laid it out in plain English: “Our AI content strategy has started a ‘doom loop’ that will hurt the performance of our models and the entire web at the same time.”

Let me break down what that actually means, because it’s worse than it sounds.

These AI models – the ones churning out articles, designs, code, whatever – they’re trained on human work. Journalists, graphic designers, photographers, writers. Real people making real things. But here’s the sick irony: the better these models get at mimicking that work, the less companies need to pay actual humans to do it. Which means less human-created content gets made. Which means the AI has less quality stuff to learn from. Which means the models get worse over time.

It’s like… imagine a restaurant that makes amazing soup by buying vegetables from local farmers. Then the restaurant decides to undercut the farmers so badly they go out of business. Suddenly there’s no fresh vegetables anymore, and your soup turns to garbage. Except in this case, the restaurant saw this coming and decided the short-term profits were worth it anyway.

The Part That Really Gets Me

One Microsoft document actually said this was “highly unusual” – that an end product would threaten the “economic foundations” of its own suppliers. Highly unusual. That’s the corporate speak for “we know this is insane but we’re doing it anyway.”

And look, I’ve covered tech for fifteen years. I’ve seen companies make ruthless calls before. But this isn’t just ruthless. It’s self-destructive. It’s burning down the library to heat your house for winter.

Who Actually Loses Here?

Not the executives, that’s for sure. They’ll get their bonuses for “innovation” and move on before the consequences really hit. The people who lose are the ones creating content right now – and I mean everyone from freelance writers to photographers to the person making infographics for corporate blogs.

But it’s not just about jobs (though yeah, it’s definitely about jobs). It’s about what happens to the web itself. Right now, if you search for basically anything, you’re starting to see AI-generated slop clogging up the results. Articles that sound fine but say nothing. Images that look okay until you notice the hands have seven fingers. It’s like the internet is slowly filling up with junk food – technically edible, but it’ll rot your brain if that’s all you consume.

And the thing is, once this content saturates the web, the next generation of AI models trains on… what? Their own output? It’s called model collapse, and researchers have been warning about it for a while now. You end up with AI training on AI training on AI, and the quality degrades with each generation. Like making a photocopy of a photocopy of a photocopy.

“Our AI content strategy has started a ‘doom loop’ that will hurt the performance of our models and the entire web at the same time.”

The Backlash Is Building

What’s actually encouraging – if anything about this mess can be encouraging – is that people are starting to push back. The lawsuit that exposed these documents in the first place is part of a wider pattern. Media companies, artists, writers, they’re not just rolling over anymore. And inside the AI companies themselves, engineers and executives are starting to say publicly what they’ve apparently been saying in private meetings: this might be a really bad idea.

The disclosure of Microsoft’s internal warnings came right as several top AI executives have been issuing public warnings about the technology they’re building. Which is wild when you think about it. It’s like tobacco executives in the 1960s standing up and saying “hey, these cigarettes we’re selling might actually kill you.” Except they kept selling them anyway.

What This Actually Means

Here’s what keeps me up at night about this: we’re watching it happen in real time, and I’m not sure we can stop it. The incentives are too strong. Microsoft, OpenAI, Google, they’re all racing to dominate AI, and nobody wants to be the one who slows down and gets left behind. Classic tragedy of the commons stuff – everyone acting in their own short-term interest, destroying the shared resource. The AI Gold Rush That Doesn't Add Up becomes a lot clearer when you see internal teams documenting their own collapse in spreadsheets.

The web as we know it took decades to build. Millions of people contributing knowledge, creativity, expertise, weirdness, passion. And we might be watching a handful of companies hollow it out in just a few years, fully aware that’s what they’re doing, because the quarterly earnings look good.

I don’t have a neat solution here. Maybe the lawsuits work. Maybe regulation catches up (though let’s be honest, that usually takes forever). Maybe the model collapse happens fast enough that these companies realize they’ve killed their golden goose before it’s too late. Or maybe we’re just going to have to rebuild from scratch, creating new spaces that explicitly exclude AI slop, like some kind of digital speakeasy for human-made content. The Great AI Bet: Why Banks Are Already Bailing Out suggests some institutions are reading the same warning signs Microsoft's own people flagged months ago.

What I do know is that when the people building this technology are internally calling it a “doom loop,” maybe – just maybe – we should listen.