• Meta (META)'s separation from Virtue AI marks an unexpected end to the June 2026 acqui-hire that brought co-founders Dawn Song, Bo Li, and Sanmi Koyejo into Meta Superintelligence Labs and FAIR.
  • The precise scope of the split—individual departures or a broader unwinding—remains unclear, and sources say the original deal was a talent recruitment push, not a conventional acquisition.
  • The development raises fresh questions about Meta's ability to integrate elite AI-safety expertise as it pours up to $145 billion into AI infrastructure and navigates active EU AI Act enforcement.

A Talent Bet Unravels

Meta's carefully orchestrated acqui-hire of Virtue AI's leadership has hit a snag, according to a Semafor report that sent ripples through the AI talent market. The June arrangement, which brought co-founders Dawn Song, Bo Li, and Sanmi Koyejo—along with several other staff—into Meta's AI organizations, now appears to have substantially changed or ended, according to people familiar with the matter.

The precise contours of the split were not immediately detailed in public reporting, leaving open the question of whether the founders are departing individually, whether a broader team exit is underway, or whether some product or support arrangement has been terminated. That distinction matters: prior reporting characterized the original transaction as an acqui-hire rather than an acquisition of Virtue AI's corporate entity, IP, contracts, or entire workforce. Meta reportedly did not purchase the startup outright.

Song and Li had joined Meta Superintelligence Labs, while Koyejo went to FAIR, Meta's fundamental-AI research arm, according to prior reports. Terms were never disclosed. Represented by leaders like Nat Friedman on the products side and Rob Fergus at FAIR, the hires were meant to harden Meta's frontier-AI effort with specialized red-teaming and security expertise.

Meta did not respond to a request for comment by publication time. Virtue AI could not be immediately reached.

A Costly Separation

If the departure weakens Meta's safety bench, it comes at an inopportune moment. The company is spending heavily to internalize AI security and model-evaluation capabilities as it deploys increasingly autonomous AI agents and tools that can call APIs, write code, and act on behalf of users—applications that demand robust safeguards against data exfiltration, prompt injection, and unsafe tool use.

The separation also lands amid broader organizational turbulence. Meta cut roughly 10% of staff in May 2026 and has pursued "Project OT," an initiative to make the workforce more "AI native" through smaller teams and less management, according to Reuters. The company reportedly scrapped a planned second wave of cuts after internal backlash and concerns about the reliability, security, and productivity of internal AI agents and coding tools.

Holding or hiring leading AI researchers does not automatically translate into durable organizational integration, dependable internal systems, or faster execution. The Virtue AI reversal may reinforce that lesson.

Financial pressure adds to the complexity. Meta reported Q2 2026 revenue of $60.80 billion, up 28% year over year, with advertising revenue rising 27% to $59.36 billion. But net income fell 14% to $15.85 billion, and operating margin dropped to 31% from 43%, reflecting sharply higher spending. Q2 capital expenditures hit $31.08 billion, and full-year 2026 capex is forecast at $130 billion to $145 billion, largely for AI infrastructure. Free cash flow was just $784 million in the quarter despite $31.86 billion in operating cash flow.

Regulatory Stakes Rise

The timing is delicate. The European Commission began enforcing AI Act provisions on August 2, 2026, including transparency obligations for certain AI systems—informing users when they are interacting with AI and labeling AI-generated or altered content such as deepfakes. Those requirements increase the commercial value of safety, testing, traceability, and compliance expertise—the precise capabilities Virtue AI was built to provide.

A Meta–Virtue split may therefore heighten questions about continuity of compliance expertise as Meta expands AI products internationally. The company also faces ongoing U.S. competition-policy risk, with the FTC seeking to revive its challenge to Meta's Instagram and WhatsApp acquisitions.

Virtue AI was founded in 2024 by prominent academics and AI-security researchers, including Song, Li, Koyejo, and Carlos Guestrin. It raised $30 million in 2025 and offered AI red-teaming and compliance tools to enterprise customers. If the startup re-emerges independently, it could become a neutral supplier of safety infrastructure to multiple model providers—a role that may grow in value as AI Act enforcement and enterprise governance demands intensify.

The larger trend is unlikely to reverse: AI security, red-teaming, and compliance are becoming core product and infrastructure functions rather than peripheral research specialties. Meta's financial capacity remains formidable, but its ability to turn exceptional infrastructure spending and elite talent acquisition into commercially reliable AI products—without intensifying regulatory, safety, or labor backlash—is the strategic test.

Correction: An earlier version of this article misstated the placement of one of the Virtue AI co-founders. Sanmi Koyejo joined FAIR, Meta's fundamental-AI research organization.