Meta parts ways with Virtue AI team after just four months

In June 2026, Song Xiaodong announced on social media that she was joining Meta Superintelligence Labs. According to media reports at the time, the other two Virtue AI founders, Bo Li and Sanmi Koyejo, joined at the same time along with more team members.

It is worth noting that this was Meta bringing in an independent team, not acquiring the entire Virtue AI company. This point defines the boundaries of the subsequent discussion: it was more a formation-level talent acquisition than a merger integration.

Song Xiaodong's academic weight is one reason this hire drew attention. The Machine Heart article mentions that her 2005 paper proposing "Dynamic Taint Analysis" is a classic paper in the field of computer security; on the AMiner ranking, she is the world's most-cited scholar in computer security; and her lab at Berkeley has been called the Whampoa Military Academy of the computer security field. Beyond her professorship, she is also the founder of Oasis Labs and Virtue AI.

According to Virtue AI's official website, its business includes simulating attacks, finding AI's weak points, and providing protection at runtime while implementing security and governance rules. In plain terms, it both helps AI get a "security checkup" and installs "guardrails" for it during operation.

In a recent article, foreign media outlet Semafor wrote that Meta spokesperson Andy Stone told Semafor the company is cutting a group of employees who were hired from the AI security startup Virtue AI and who joined in June of this year, only four months ago.

As for why it reached this point so quickly, Meta's explanation is "clashing work styles," saying the collaboration did not develop as expected. Meta also said Superintelligence Labs will still focus on AI safety, alignment, and frontier risks.

What needs to be distinguished is this: the action of cutting this group of employees was confirmed by a Meta spokesperson to Semafor; while "clashing work styles" is an explanation given by Meta's side, a company statement, and the specific disagreements were not disclosed.

In addition, the report also did not confirm whether Song Xiaodong and the others all left, and the team had not responded as of the report's publication. The specific number of employees cut was likewise not given in the sources.

The Machine Heart article argues that the explanation of "clashing work styles" clearly fails to convince many people, and what deep disagreements existed between the two sides still awaits an explanation from Virtue AI.

This dissatisfaction has a structural reason: based on public information, when the two sides came together in June, the logic was complementary—the Virtue AI team was already doing AI security testing, red teaming, and post-launch protection for clients, and as a startup it had limited scenarios and resources, so entering Meta could scale these capabilities into much larger products; while Meta was aggressively pushing AI Agents and superintelligence and needed a team that understood both research and security implementation to backstop AI safety, reliability, and governance.

But the above complementary logic is the Machine Heart article's account of the motivations at the time, not an official statement confirmed by both sides. Therefore, it can only explain "why it seemed reasonable at the beginning," not "why it ended after four months."

The truly unanswered question is: if the disagreement was not about capability, and not about direction, then where exactly did it occur? Existing sources do not provide an answer.

The Machine Heart article introduces an external study as a reference point. A study published by MIT Sloan in 2019 found that within the first year of a company being acquired, 33% of acquired employees chose to leave, compared with 12% of regularly hired employees with similar skills and work experience.

The study also said that over time these percentages tend to converge, but within the three-year window it studied, acquired employees were 15% more likely to leave than new hires.

The article argues that the core reason for the high attrition rate of acquired employees is "organizational mismatch": these people joined a startup in the first place for the entrepreneurial atmosphere, autonomy, flexibility, and risk-taking environment, but after being acquired by a large company they often face more bureaucratic processes and a formal corporate culture, which directly conflicts with their preferences; moreover, ordinary acquired employees usually cannot choose the buyer and cannot participate in post-acquisition organizational adjustments, and this "no choice" further intensifies the discomfort.

Two limitations must be flagged here. First, this is the Machine Heart article's inference, not the literal conclusion of the MIT Sloan study. Second, the MIT Sloan data was published in 2019 and concerns ordinary acquired employees, whereas this incident involves Meta bringing in an independent team rather than acquiring a company, so the situations do not fully correspond. Therefore this set of numbers can only serve as a reference for understanding "why talent tends to drain away after being brought in," and cannot be directly applied to Song Xiaodong's team.

The Machine Heart article proposes that the core question truly left behind by this collaboration change is: after bringing excellent researchers in, how do you let them contribute within the organization?

This question matters more than "who was right and who was wrong" because it does not depend on the specific attribution of this incident. Even if "clashing work styles" is the real reason, it describes only the result, not the mechanism—how an organization accommodates a team brought in as a formation, carrying its own way of working and security philosophy.

Based on existing sources, it is not clear whether Meta Superintelligence Labs' subsequent arrangements for AI safety, alignment, and frontier risks are affected; the sources also do not explain the direction and status of the Virtue AI company itself after the collaboration change.

Also left hanging: whether "clashing work styles" is the real reason or an external talking point cannot be judged from existing sources. Whether Song Xiaodong herself and team members such as Bo Li and Sanmi Koyejo all left also remains unclear.