Meta Parts Ways with Virtue AI Co-Founders Four Months After Hiring Team

Meta has parted ways with several employees recruited from AI safety startup Virtue AI, including its co-founders, approximately four months after bringing the team on board to strengthen its artificial intelligence safety and security capabilities.

The departures include Virtue AI co-founders Bo Li, Dawn Song and Sanmi Koyejo, who joined Meta in June alongside other members of the startup’s team. The recruitment was viewed as part of Meta’s efforts to expand its expertise in AI safety as competition for specialised researchers and engineers intensifies across the technology industry.

Meta spokesperson Andy Stone reportedly said the arrangement had not worked out as planned, citing differences in working styles. The company has maintained that Meta Superintelligence Labs continues to focus on AI safety, alignment and research into risks associated with advanced AI systems.

Virtue AI Developed AI Security and Governance Tools

Virtue AI specialised in tools designed to evaluate and strengthen the security of artificial intelligence systems. Its work included automated red-teaming technologies, which test AI models for vulnerabilities, and runtime safeguards intended to identify or prevent potentially harmful behaviour.

The startup had also worked with organisations including Anthropic and OpenAI, as well as the US National Institute of Standards and Technology (NIST).

Meta recruited members of the Virtue AI team rather than acquiring the startup outright. The subsequent departures highlight the potential challenges involved in integrating specialised teams into large technology organisations, particularly when employees bring established research practices and working methods.

AI Talent Integration Remains a Challenge for Technology Companies

The short tenure of the former Virtue AI team members raises questions about how technology companies integrate specialist expertise into existing research structures.

For HR leaders, the development underscores the importance of cultural alignment, leadership expectations and clear responsibilities when hiring entire teams or recruiting senior researchers from startups.

Acqui-hiring and team-based recruitment can help organisations secure specialised talent quickly. However, differences in decision-making processes, research priorities and working styles can complicate integration if expectations are not aligned from the outset.

Companies competing for AI safety professionals may therefore need to focus not only on attracting talent but also on creating working environments in which specialist teams can collaborate effectively and contribute to long-term research goals.

The departures come amid continuing discussions in the United States about oversight of advanced AI systems and the responsibilities of companies developing increasingly capable models.

Meta has said the changes do not represent a shift in its broader commitment to AI safety and frontier-risk research.