Meta is reorganizing its AI division for the fourth time in six months, splitting the months-old Meta Superintelligence Labs (MSL) into four teams and imposing a division-wide hiring freeze, according to an internal memo from new chief AI officer Alexandr Wang.

The new structure places Wang at the helm of TBD Lab, a small group focused on frontier models and the Llama series. FAIR, Meta’s long-standing research lab now run day-to-day by research director Rob Fergus, is repositioned to feed longer-horizon work into TBD Lab’s model training. A Products and Applied Research group, led by former GitHub CEO Nat Friedman, will fold AI into consumer products, while MSL Infra, under engineering VP Aparna Ramini, handles compute and data centers. The AGI Foundations team created in May is dissolved, and its members are dispersed across MSL’s product, infrastructure and FAIR divisions. Chief AI scientist Yann LeCun now reports to Wang too; the one senior name the memo doesn’t place under him is ChatGPT co-creator Shengjia Zhao, installed as MSL’s chief research scientist.

The restructuring comes after a summer of aggressive talent raids that saw Meta poach more than 50 researchers from rivals, some on nine-figure pay packages. The New York Times reports that Meta’s stalled “Behemoth” frontier model is being set aside for a fresh approach, and that the company is weighing whether to keep its next big model closed rather than open-weight. The freeze halts outside hires and internal transfers unless Wang personally signs off.

The moves have intensified tensions between new high-profile hires and veteran researchers, some of whom have threatened to quit. Critics see the repeated shakeups as signs of panic; Wang’s memo describes them as necessary to get the structure right for long-term velocity toward superintelligence.