Published: October 3, 2026 · Last updated: October 3, 2026
TL;DR: In June 2026, Meta pulled three co-founders of AI security startup Virtue AI into its Superintelligence Labs and FAIR research group. Four months later, according to a Semafor report, Meta and Virtue AI have split, and nobody on either side will say why.
Acqui-hires are supposed to be the boring, easy kind of AI deal. A startup’s talent walks in the door, the org chart absorbs them, and six months later nobody remembers there was ever a separate company. That is not what happened with Virtue AI.
According to a Semafor report that circulated widely on October 2, Meta has parted ways with the team it hired out of Virtue AI only four months earlier. The original deal, reached in June 2026, brought three of the startup’s co-founders, Dawn Song, Bo Li, and Sanmi Koyejo, into Meta’s orbit. Song and Li joined Meta Superintelligence Labs. Koyejo went to FAIR, Meta’s fundamental AI research group. A fourth Virtue AI co-founder, Carlos Guestrin, never joined Meta at all and appears to still be associated with whatever is left of the startup.
Neither Meta nor Virtue AI would confirm what actually happened. Meta did not respond to requests for comment before the Semafor story published. Virtue AI could not be reached. That silence is doing a lot of the talking here: when a company declines to explain why it just unwound a high-profile AI safety hire, the simplest explanation is rarely the generous one.
What makes the reversal sting is the framing of the original deal. People briefed on it at the time described it as a talent push, not a corporate acquisition. Meta reportedly never bought Virtue AI’s actual company, its intellectual property, its contracts, or its full workforce. The terms were never disclosed publicly, and the arrangement left Virtue AI technically independent even as its most recognizable names moved into Meta’s buildings. Nat Friedman represented the product side of the integration; Rob Fergus represented FAIR. Several other staff came along in the original move, though nobody has said how many, and Semafor’s reporting stops short of saying whether this is a handful of individual departures or a wider unwinding of the whole arrangement.
Virtue AI itself is not a nobody in this space. Founded in 2024 by Song, Li, Koyejo, and Guestrin, four people with real academic standing in AI security research, the startup built red-teaming and compliance tools that enterprise customers used to stress-test their own models before shipping them. It raised $30 million in 2025. If the founders end up back at the company full time, there’s a real argument that Virtue AI re-emerges stronger: a neutral safety vendor that isn’t tied to any single model maker’s incentives is arguably a better pitch than a safety team embedded inside the lab whose products it’s supposed to be checking.
Meta’s own numbers give some sense of the pressure the company is under while this plays out. It is pouring somewhere between $130 billion and $145 billion into AI infrastructure this year. Q2 revenue came in at $60.8 billion, up 28% year over year, but net income fell 14% to $15.85 billion and operating margin dropped from 43% to 31%. Meta cut around 10% of its staff back in May under an initiative called Project OT, aimed at making teams smaller and more “AI native,” and reportedly scrapped a second round of cuts after employees raised concerns about how reliable its internal AI coding tools actually were. None of that proves anything about why the Virtue AI team left. It does explain why a company spending this kind of money on AI, while absorbing this kind of internal churn, might struggle to keep an acquired team settled.
There’s also a regulatory backdrop that makes the timing awkward. The European Commission started enforcing new AI Act transparency rules on August 2, requiring companies to disclose when users are interacting with AI and to label synthetic content. Meta is separately fighting the FTC’s attempt to revive its antitrust case over the Instagram and WhatsApp acquisitions. A company under that much scrutiny generally wants its AI safety story to look stable, not like a revolving door.
Related: OpenAI Fired Three Safety Researchers Over a Leak It Still Won’t Explain and A Tech Company Owner Is Accused of Smuggling $300 Million in Nvidia Chips to China.
Bottom Line: Holding onto brilliant AI researchers is not the same thing as building a system that can actually use them. Meta just proved that the hard way, twice in four months, and the explanation it owes the people who track this stuff is still sitting unanswered.
Follow Teck Hustlers: Facebook · X · Instagram · LinkedIn · TikTok · Pinterest
