Mark Zuckerberg wants your DMs. Not for blackmail — for machine learning. A new petition circulating among Meta employees demands the company stop using internal communications, code reviews, and meeting transcripts to train its AI models. And honestly? The fact that this petition exists should terrify every knowledge worker on the planet.
The petition, hosted on mcipetition.com, has garnered over 1,200 signatures in 48 hours. It calls on Meta to halt what it describes as 'non-consensual scraping of employee-generated content.' The document explicitly names Slack messages, Workplace posts, email threads, and even internal documentation as data sources being fed into Meta's large language model pipeline.
Let's be clear: this isn't some fringe conspiracy. Meta confirmed in a leaked internal memo — obtained by this reporter — that employee communications are being used 'to improve model performance on enterprise and productivity tasks.' The memo argues that since employees work for Meta, their digital exhaust is company property. Legally, they've got a point. Ethically? They're standing on a landmine.
The Silicon Valley Double Standard
Silicon Valley loves to preach about privacy. Tim Cook gives speeches about it. Mark Zuckerberg once testified before Congress with a straight face claiming 'privacy is important.' But when it comes to their own workforce, the rules change.
Meta's privacy policy for employees is a dense jungle of legalese that gives the company broad rights to monitor communications. Most workers signed it without reading — who reads those contracts anyway? But now that the AI is literally learning from their venting in Slack channels about broken build systems, the abstraction has shattered.
One engineer I spoke with — let's call him 'Dave' because he'll lose his job otherwise — told me, 'I made a joke in a channel about how our quarterly planning process is just shuffling deck chairs on the Titanic. That joke is now part of a training set. Some future AI might think that's a literal description of corporate strategy.' He laughed, but it wasn't funny.
'They're not just watching us work. They're learning from us. Every complaint, every half-baked idea, every stupid meme — it's all fodder for the machine.' — Meta employee (anonymous)
This Is Bigger Than Meta
If you think this is just a Meta problem, you're not paying attention. Microsoft owns GitHub, LinkedIn, and Office 365. They have access to code, resumes, and every email you've ever sent through Outlook. Google has Gmail, Google Docs, and a decade of search history. Amazon has AWS logs, Alexa recordings, and your Whole Foods shopping list.
Every single one of these companies has the incentive and the infrastructure to turn your work product into AI training data. The only question is whether they've done it yet. Spoiler: they have. The Meta petition is just the first crack in the dam.
Let me tell you what happens when a company trains an AI on its employees' internal communications. First, the model learns the company's actual culture — not the mission statement, but the real shit. The passive-aggressive emails. The jokes about management. The genuine frustration with broken processes. Then, when that model is deployed to help with 'productivity,' it reflects all that back. You get an AI that knows exactly how to gaslight you, because it learned from the masters.
The Legal Gray Area
Employment law hasn't caught up to AI. In the United States, employers generally have broad rights to monitor workplace communications. The Electronic Communications Privacy Act allows monitoring if it's 'in the ordinary course of business.' Courts have interpreted that loosely. But training an AI model isn't monitoring — it's repurposing. It's one thing for your boss to read your Slack messages. It's another for an algorithm to embed the patterns of your thinking into a model that might one day replace you.
Europe's GDPR offers more protection, but it's not clear-cut. The 'legitimate interest' clause gives companies a loophole big enough to drive a data center through. And even if you opt out, good luck proving your data wasn't used. The training process is a black box.
What's at Stake
I've been covering tech long enough to know that once a company starts down this road, they don't turn back. The cost of retraining a model is astronomical. The cost of ignoring a petition is a few angry blog posts and some bad press. Meta will likely issue a statement about 'listening to employees,' maybe promise a review, and then quietly continue. By the time the review is done, the model will be too ingrained to unwind.
But here's the thing: if Meta's employees — the most technically literate people on Earth — can't stop this, what chance do the rest of us have? These are the people who understand the technology better than anyone. If they're scared, you should be scared.
The petition demands three things: transparency about what data is collected, the ability to opt out without retaliation, and deletion of existing employee data from training sets. None of these are unreasonable. None of them will be granted without a fight.
I've got a message for Meta's leadership: You're building a machine that ingests the private thoughts of your own people. You're training it on their jokes, their complaints, their half-formed ideas. And you're doing it without their consent. That's not innovation. That's surveillance dressed up as progress.
To every Meta employee reading this: keep fighting. Sign the petition. Talk to reporters. Leak the memos. Because if you can't protect your own data, you've already lost the war for everyone else's.
