Apple’s effort to build its own large language models is under strain as key researchers depart and management wrestles over strategy, according to a report by The Information corroborated by subsequent media coverage. The roughly 40-person foundational models group has seen a handful of exits in recent months, with head Ruoming Pang among those said to be weighing outside offers. Tensions flared earlier this year when the team sought to release several of its models as open-source software to showcase progress and attract external collaboration. Craig Federighi, Apple’s senior vice president of software engineering, vetoed the plan, warning that weaker on-device versions could damage the company’s image if their performance lagged rivals’ cloud-based systems. Morale slipped further after Apple postponed a new Siri powered by in-house models to 2026 and began discussing the use of third-party systems from OpenAI, Anthropic and Google. Researchers complained they were not consulted on the delay, undermining confidence in senior leadership. Competitors are exploiting the turmoil: Meta has aggressively recruited Apple scientists, reportedly dangling compensation packages of about $200 million for high-profile hires such as Pang. While AppleInsider notes the unit is unlikely to unravel overnight, the departures underscore the fierce talent war engulfing advanced AI research. The internal discord highlights the challenge Apple faces in balancing its privacy-focused, on-device AI philosophy with the rapid, cloud-centric advances achieved by rivals. How the company responds could determine whether it can close the widening gap in generative AI capabilities.