Meta breaks up its Responsible AI team
Meta has disbanded its Responsible AI (RAI) team, reassigning most members to its generative AI product team and others to AI infrastructure. Meta insisted it will continue prioritizing safety, but the move mirrors Microsoft's earlier cuts. Why it matters: In the heat of the generative AI race, centralized ethics oversight is rapidly getting swallowed by product deadlines.
Big Tech loves talking about AI ethics, right up until those ethics slow down product shipping.
Meta has officially disbanded its Responsible AI (RAI) team, according to an internal post first reported by The Information. Most of the team's employees are being sent to Meta’s generative AI product division, while the rest will move over to AI infrastructure.
Meta spokesperson Nisha Deo confirmed the shakeup to The Verge, framing it as a way to better support AI feature development. Deo insisted the company will keep investing in safe development, claiming former RAI staff will continue to support cross-company safety work from their new desks.
A safety team that was already running on empty
This breakup didn't happen in a vacuum. It looks more like the final nail in the coffin.
Meta set up the RAI team back in 2019 to catch systemic issues before they hit the real world, such as ensuring diverse training data and preventing toxic platform moderation slips. But earlier this year, Business Insider reported that previous layoffs had already gutted RAI into a mere shell of a team.
Insiders noted that the group lacked real autonomy. Pushing through even basic safety initiatives required endless, grueling negotiations with business stakeholders.
Meanwhile, Meta's automated track record is messy. Over the years, translation glitches on Facebook led to a wrongful arrest, WhatsApp's AI sticker tool pumped out biased images, and Instagram’s algorithms surfaced child abuse material. Disbanding the watchdog doesn't mean those problems disappeared.
Why it matters
Moving safety researchers into product teams sounds great in a corporate memo, but in practice, it usually means safety takes a backseat to shipping speed.
We think this decentralized approach is risky. When safety auditors report directly to the product managers whose bonuses depend on launching generative features yesterday, the incentive to sound the alarm drops to zero. A watchdog inside the sled team rarely pulls the brakes.
Meta isn't alone here. Microsoft gutted its own Ethics and Society group earlier this year under similar restructuring pretexts. Across the board, frontier labs are clearly treating centralized safety teams as expensive speed bumps in the race to catch OpenAI.
It also seems remarkably short-sighted. Regulators are circling fast, with the White House rolling out an executive order on AI safety and the European Union trying to finalize its AI Act. Dismantling the exact group meant to keep models compliant leaves Meta wide open to massive regulatory blowback the next time its algorithms misbehave.
Sources (1)
- theverge.com(opens in a new tab)Originaltheverge.com