Cognitive sonars dynamically tune system parameters to improve performance in pursuit of specific goals. Most research on cognitive sonar focuses on active sonar, varying the transmitted waveform and ping rate. Although passive sonar also faces the challenges of pursuing goals in varying and unknown environments, little research has been done exploring cognitive approaches to passive beamforming for sonar arrays. Universal beamformers provide one implementation of a cognitive approach to passive sonar. Practical adaptive beamformers generally regularize the sample covariance matrix before estimating the array weights. Universal beamformers blend the array weights across a family of beamformers competing on different choices for the regularization parameters. The blend of array weights is performance driven based, including the largest portion from the beamformers best suited for the current environment. Universal are “doubly adaptive” in that each competing beamformer is adapting its array weight vector in response to the data observed at the sensor array, the universal algorithm is then meta-adapting the blend of these array weight vectors used to process the data based on the performance of each competing beamformer. We will present examples of beamformers, which are universal over dominant subspace dimension and beam pattern notch width for moving interferers. [Work supported by ONR 321US.]