AdaJEPA, a novel adaptive latent world model, has been introduced to enhance planning from high-dimensional observations by predicting future states in a compact latent space. This approach addresses the limitations of traditional latent world models, which often fail when their predictions become inaccurate, particularly under test-time distribution shift. AdaJEPA's adaptive nature allows it to refine its predictions at test time, improving planning outcomes. By performing test-time adaptation, AdaJEPA can better handle changes in the environment or distribution shifts, making it a more robust and reliable option1. The implications of this development extend beyond the realm of artificial intelligence, as state-aligned threat activity can raise the stakes from criminal to geopolitical. As a result, the ability to adapt and respond to changing circumstances becomes crucial. This matters to practitioners because adaptive models like AdaJEPA can help mitigate the risks associated with distribution shifts, ultimately leading to more effective and resilient planning systems.