Meta tells “drafted” engineers for Applied AI they can opt out, “defer to” choice

Meta is giving engineers in its Applied AI training unit an escape hatch. According to an internal memo obtained by Business Insider and four people familiar with the matter, the company sent a Wednesday note telling employees who were “drafted” into the unit that it will now “defer to each individual's choice.”

This matters because the Applied AI push was not framed as voluntary. Last month, Meta reassigned 7,000 employees to units including an Applied AI task force, aimed at helping train Meta’s coming AI models. The memo makes the reversal explicit: Meta “prefer[s] everyone to stay and push to SOTA together,” referring to state-of-the-art, but then adds that the company will support employees in whatever decisions they make and “defer[s] to each individual's choice.”

So what changed between last month’s scramble and this week’s opt-out memo? The source story connects the unwind to a morale reckoning inside Meta. Business Insider previously reported that on June 2, during an internal “Tuesdays with Boz” session, Meta CTO Andrew Bosworth told employees morale was “probably one of the worst it's ever been” in Meta’s 20-year history. Meta’s Applied AI task force drew significant backlash last month, with employees comparing the work to data labeling. The contrast between “drafted” and “undraft,” as some employees called it on Blind, is basically the whole plot.

There’s also a practical staffing angle baked into the memo. People in the Applied AI unit would receive preferential placement in other parts of the company due to staffing shortages. In other words, Meta is not just saying “you can leave.” It is also signaling that it wants the unit to keep talent where possible, but it is no longer willing to fight every individual battle over agency and assignment. That’s a subtle but important shift in how a company communicates power: you can still “prefer everyone to stay,” but you stop treating refusal as a problem to solve.

This is happening in the middle of Meta’s broader workforce turbulence. In May, Meta laid off 10% of its staff, or 8,000 people. That context matters because even when companies talk about “building AI,” employees experience reorganizations like this through the lens of job security, workload, and recognition. Training models requires huge amounts of human time, judgment, and iteration. But when Meta’s initial framing sounded like engineers were reassigned by default, the optics looked less like upskilling and more like compelled labor, especially to employees already sensitized by layoffs.

From a risk-management perspective, the reversal also reduces friction that can metastasize. In most engineering organizations, morale dips are not just “feelings,” they show up as attrition risk, slower execution, and more internal drag. When a task force becomes synonymous with an analogy like data labeling, you get narrative damage on top of operational change. The memo’s tone aims to defuse that narrative by explicitly restoring personal agency, and by tying employee decisions to internal placement rather than leaving people to wonder if opting out is career suicide.

For executives and boards at other tech firms, this is a useful stress test. Meta is publicly known for aggressive execution, but the internal governance lesson here is about how quickly employee alignment can flip when directives are perceived as coercive. It is also a reminder that AI initiatives are not only technical programs; they are also workforce programs. If the talent story breaks, the training pipeline suffers, whether you measure that in retention, speed, or the quality of the work.

Meta declined to comment for this story. Still, the internal memo’s language is clear about the immediate outcome: Applied AI is no longer a one-way “draft.” Now it is a choice. And for leaders who are planning similar internal AI reorganizations, the message is hard to ignore: you can mobilize for SOTA, but when morale is already fragile, “prefer everyone to stay” is not the same as “defer to each individual's choice.”