{"slug": "its-literally-the-gulag-inside-the-revolt-at-metas-ai-unit-where-elite-engineers", "title": "‘It’s literally the gulag’: inside the revolt at Meta’s AI unit, where elite engineers were drafted to label data", "summary": "Meta's Applied AI unit, a 6,500-person team created to power CEO Mark Zuckerberg's AI ambitions, is in turmoil after elite engineers were forcibly reassigned to data-labeling tasks, with employees describing the environment as \"the gulag.\" The unrest follows a $14.3 billion investment in Scale AI, layoffs of 8,000 workers, and a controversial keystroke-tracking system, culminating in a major Instagram security breach and the departure of Meta's chief information security officer.", "body_md": "It takes a particular kind of dysfunction for an employee to hijack a company-wide livestream and demand the hosts pass a message to a senior executive: that he is, in not so many words, garbage.\n\nThat happened inside Meta this month. According to a recording heard by WIRED, the outburst, on a call open to thousands of staff, was not really about one executive. It was the loudest signal yet of how badly things have soured inside Applied AI, the roughly 6,500-person unit Mark Zuckerberg built in March to power his most expensive bet in artificial intelligence.\n\n“It’s literally the gulag,” one employee told WIRED. “You have zero purpose in life all of a sudden.”\n\n## The $14.3bn bet, and the machinery behind it\n\nThat bet has a name: Alexandr Wang. Last summer, Meta paid $14.3bn for a 49 per cent stake in Wang’s data-labelling startup, Scale AI, and installed the young founder as its chief AI officer, leading the new Meta Superintelligence Labs. Applied AI is the machinery meant to turn his models from promising into competitive with Claude and ChatGPT.\n\nThe problem is what that machinery actually does. Applied AI is led by Maher Saba, a 12-year Meta veteran and former Reality Labs vice-president, reporting up to chief technology officer Andrew Bosworth.\n\nAs The Pragmatic Engineer’s Gergely Orosz documented in a detailed account this week, between 30 and 50 per cent of engineers on core product, infrastructure and security teams were reassigned to an internal group focused on data labelling and reinforcement learning from human feedback, the human-in-the-loop grunt work that improves AI models.\n\nAround 6,500 people now sit in it; by his estimate, one in every five or six of Meta’s engineers may be labelling data full time.\n\nFor most of Meta’s history, engineers chose their own teams. Now, per an internal memo from the unit’s chief seen by Reuters, the transfers were not optional. The people moved against their will began calling themselves draftees.\n\n## Surveillance, layoffs and tokenmaxxing\n\nNone of this happened in a vacuum. In May, Meta [cut about 8,000 jobs](https://thenextweb.com/news/meta-layoffs-8000-ai-restructuring-may-2026), a tenth of its workforce, days after shifting thousands more onto AI teams. Around the same time, staff [protested a system that tracks their keystrokes and mouse clicks](https://thenextweb.com/news/meta-mouse-tracking-protest-layoffs) to harvest AI training data, with no opt-out; more than 1,600 signed a petition, and one office nicknamed it the “Employee Data Extraction Factory”.\n\nAfter the backlash, Reuters reported on 2 June that Meta had begun letting staff pause the tracking for up to 30 minutes at a time and request exemptions.\n\nThen there were the incentives. Engineers learned that token usage would be weighed in performance reviews, so they began burning tokens for the sake of it, a status game the industry came to call [tokenmaxxing](https://thenextweb.com/news/tokenminimizing-companies-cap-employee-ai-spending). Per The Information, Meta staff ran through 60.2 trillion AI tokens in 30 days, which at list prices would cost hundreds of millions of dollars.\n\nThe perverse result, Orosz argues: an outage caused by sloppy AI-generated code might not get you fired, but writing code by hand could.\n\n## The bill comes due\n\nThat culture has a cost. On 30 May, Meta suffered what Orosz calls the most embarrassing outage in its history: a wave of Instagram account takeovers, including high-profile accounts, after its support AI could reportedly be tricked into sending a password-reset code to an attacker’s email.\n\nMeta’s security teams, gutted by reassignments, were caught flat-footed; its chief information security officer departed days later. Meta has separately had to freeze some [AI data work after a breach put training secrets at risk](https://thenextweb.com/news/meta-mercor-breach-ai-training-secrets-risk).\n\nEven leadership has stopped pretending. CTO Andrew Bosworth told staff the AI reorg was “atrocious” and that morale was near the worst he had seen in 20 years, comparable to Cambridge Analytica.\n\nChief product officer Chris Cox described the “insanity of this company” and likened the period to “running a marathon in the middle of a hailstorm”, before urging a reality check on the technology itself: AI, he said, “is neither god, nor is it the devil”.\n\nNow Zuckerberg is trying to steady the ship.\n\nIn a Friday memo he conceded the changes had “caused distress” and that “we’ve made mistakes and will almost certainly make more”. He has promised no further company-wide layoffs for the rest of 2026, moved to cap a manager-to-report ratio that had ballooned towards 50-to-one on teams such as Applied AI, raised budgets for team events, and floated a company-wide AI hackathon, an idea employees openly hated.\n\nWhether that is enough to stop the exodus is another question: Orosz reports a surge of Meta engineers signing up to interview-prep services since May.\n\n## The cautionary tale\n\nThe strangest part is that the business is booming.\n\nMeta posted $56.3bn in first-quarter revenue, up 33 per cent, with its AI-powered ad-creation tools doubling to eight million advertisers; on current trends it is on course to overtake Google as the world’s largest advertising business. Its latest in-house model landed to a muted response, yet Zuckerberg decided that building a coding AI mattered more than keeping his best engineers, or running Instagram and Facebook reliably.\n\nThere is a case that the market, and the morale story, are missing the point. As Forbes contributor Jon Markman argues, the $14.3bn did not buy a model at all; it bought a near-half stake in Scale AI, the supplier of the high-quality training data and reinforcement-learning environments that every frontier lab is now scrambling for.\n\nMeta gave Llama away for free precisely because Zuckerberg bet that models would commoditise and that the durable value would sit in the data and the distribution, the same logic that already turns Meta’s behavioural data into industry-leading ad returns. By that reading, grading Meta on whether its new model tops a benchmark is grading the wrong exam.\n\nIt is a genuine counterweight, even if it does nothing to soften the human cost inside the engineering org, or explain why a company posting record profits chose to gut the teams that produce them.\n\nMeta’s stock is down around 18 per cent over the past year, the weakest of the megacaps, as investors weigh that very question.\n\nIt is a pattern others see spreading.\n\nMitchell Hashimoto, the founder of HashiCorp, warns that entire companies are operating under a kind of “AI psychosis”, shipping fast on the assumption that agents will clean up the mess, the way Meta’s engineers dropped their quality bar and then watched high-profile accounts get hacked.\n\nMeta’s story is the most extreme version so far: a profitable giant that, in the words of one of its own product chiefs, turned its prized engineering culture into collateral damage. The talent it is shedding, Orosz notes, will be someone else’s gain.\n\n## Get the TNW newsletter\n\nGet the most important tech news in your inbox each week.", "url": "https://wpnews.pro/news/its-literally-the-gulag-inside-the-revolt-at-metas-ai-unit-where-elite-engineers", "canonical_source": "https://thenextweb.com/news/meta-applied-ai-unit-revolt-data-labeling-draftees", "published_at": "2026-06-18 15:53:21+00:00", "updated_at": "2026-06-18 17:05:57.206699+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-ethics", "ai-policy", "ai-infrastructure", "ai-safety"], "entities": ["Meta", "Mark Zuckerberg", "Scale AI", "Alexandr Wang", "Maher Saba", "Andrew Bosworth", "Instagram", "Gergely Orosz"], "alternates": {"html": "https://wpnews.pro/news/its-literally-the-gulag-inside-the-revolt-at-metas-ai-unit-where-elite-engineers", "markdown": "https://wpnews.pro/news/its-literally-the-gulag-inside-the-revolt-at-metas-ai-unit-where-elite-engineers.md", "text": "https://wpnews.pro/news/its-literally-the-gulag-inside-the-revolt-at-metas-ai-unit-where-elite-engineers.txt", "jsonld": "https://wpnews.pro/news/its-literally-the-gulag-inside-the-revolt-at-metas-ai-unit-where-elite-engineers.jsonld"}}