Meta Disbanded Its Responsible AI Team — And That Should Worry You

Meta just quietly dissolved its Responsible AI (RAI) team. According to a report from The Information that came out today, most of the RAI staff are being shuffled into the company’s generative AI product team, with a few others moving to work on AI infrastructure.

Meta’s communications rep, Nisha Deo, told The Verge this is all about helping “aid in the development of AI features” and that they’ll “continue to prioritize and invest in safe and responsible AI development.” She also said the RAI members now in the generative AI org will “continue to support relevant cross-Meta efforts.”

I’ve heard this kind of corporate reassurance before, and I’m not buying it. When you break up the dedicated team whose entire job is to catch problems before they ship, and scatter them across product teams, you’re effectively making safety a secondary concern. Product teams have deadlines and feature roadmaps. Safety teams have, well, safety.

This isn’t even the first time Meta has gutted RAI. Earlier this year, Business Insider reported that layoffs had already left the team as “a shell of a team.” The RAI team, which had been around since 2019, apparently had little autonomy from the start. Its initiatives had to go through lengthy stakeholder negotiations before anything could actually happen. So even when they did find problems, fixing them was an uphill battle.

What did RAI actually do? The team was supposed to identify problems with Meta’s AI training approaches — things like whether the models are trained on adequately diverse data, and whether they’ll cause moderation issues on its platforms. And let’s be honest, Meta’s platforms have a track record here. A Facebook translation error led to a false arrest. WhatsApp’s AI sticker generator produces biased images when you give it certain prompts. Instagram’s algorithms have been caught helping people find child sexual abuse material. These aren’t hypothetical risks; they’re recurring failures.

Meta isn’t alone in this move. Microsoft did something similar earlier this year. And it’s happening just as governments are finally starting to get serious about AI regulation. The US government has been signing agreements with AI companies, and President Biden directed agencies to come up with AI safety rules. The European Union has published its AI principles and is still trying to push through its AI Act.

The timing couldn’t be worse. Companies are racing to ship generative AI features, and the people who were supposed to be the internal watchdogs are being told to help build the products instead. That’s not a recipe for responsible development — it’s a recipe for more of the same problems we’ve already seen, just at a larger scale.

Meta still has a page on its website where it talks about its “pillars of responsible AI” — accountability, transparency, safety, privacy, and so on. But when you dismantle the team that actually worked on those pillars, the page starts to look a lot like PR fluff.

I’ll believe Meta is serious about responsible AI when it treats safety as a non-negotiable part of the product development process, not as a team that can be scattered to the winds whenever the company decides to chase the next big thing.