Meta has parted ways with the co-founders and other employees of Virtue AI, an AI security startup whose team it had recruited just four months earlier to strengthen its work around AI safety, security and frontier risks.
The development comes shortly after Meta’s aggressive push to build out its AI capabilities, including its Superintelligence Labs, and has raised questions about how the company is structuring its AI safety efforts as it races to develop increasingly capable AI systems.
Meta spokesperson Andy Stone told Semafor that the arrangement with the Virtue AI team “didn’t work out as planned,” attributing the decision to differences in working styles.
Despite the departures, Stone said Meta Superintelligence Labs remains focused on AI safety, alignment and frontier-risk research.
The company has not publicly detailed how many former Virtue AI employees are leaving Meta or whether the entire team recruited in June has been affected.
The hiring was announced in June, when Meta brought in Bo Li, Dawn Song and Sanmi Koyejo, the three co-founders of Virtue AI, along with other members of the startup’s team.
Virtue AI had been working on security and governance tools designed for enterprise AI systems. Its capabilities included automated red teaming, runtime guardrails and AI governance, areas that have become increasingly important as companies deploy more autonomous and capable AI systems.
Before joining Meta, Virtue AI had also worked with organisations including Anthropic, OpenAI and the US Commerce Department’s National Institute of Standards and Technology (NIST).
Governments, researchers and technology companies are increasingly debating how advanced AI systems should be tested, monitored and governed. Earlier this month, Meta itself published an updated Superintelligence Scaling Framework, outlining additional requirements around safety testing, model evaluation, containment and monitoring as AI capabilities become more advanced.
Meta’s framework says the company is expanding its safety requirements to cover risks that can emerge during training and evaluation, rather than only after deployment. It also outlines measures such as sandboxing high-risk environments, real-time monitoring and automated systems capable of halting certain training runs when serious risks are detected.
That makes the departure of a team specifically recruited for AI security and safety particularly noteworthy, even though Meta maintains that its broader safety and alignment work remains a priority.
The Virtue AI exits are also taking place against the backdrop of a much larger restructuring at Meta.
