Meta Chief AI Officer Alexandr Wang announced the company will no longer open-source its most advanced frontier models.
The policy shift follows safety concerns identified during the development of the Muse Spark model.
Early training of Muse Spark triggered high-risk alerts, including potential biological risks.
Meta released the powerful Muse Spark model as proprietary software in April to manage these unaddressed dangers.
Wang noted that rival AI laboratories face similar challenges as model capabilities increase.
This decision marks a significant departure from Meta’s established strategy of releasing AI models openly.