Many natural systems, including the atmosphere and ocean, are dissipative and chaotic, implying sensitivity to initial conditions and that the estimation error strongly projects on the unstable manifold of the dynamics. This latter property has inspired the development of a new class of algorithms known as assimilation in the unstable subspace (AUS) in which the span of the leading Lyapunov vectors, or a suitable approximation, is used explicitly in the update step that is confined to the unstable subspace. The AUS has been incorporated and formalized in the Gaussian framework and successfully applied to atmospheric, oceanic, and traffic models. Even in high-dimensional systems, an efficient error control is achieved by monitoring only a limited number of unstable directions. The AUS provides an idealized, albeit general framework, to investigate low dimensional approximations for uncertainty quantification. The driving questions in EmblAUS are: (1) Can we improve our usage of the EnKF and PF algorithms by considering the AUS perspective? (2) Can the AUS paradigm be applied to design new Bayesian Ensemble Filters?