Is it possible to efficiently convert any arbitrary source with sufficiently high (min-)entropy into nearly uniformly random bits? Information-theoretically, this is achievable using seeded extractors, provided the source is independent of the seed. However, in the absence of any such independence guarantee, no solution is possible for general computationally unbounded sources. Even for efficiently samplable sources, we cannot extract randomness that is statistically close to uniform, but can at best condense the randomness into an output that is only missing logarithmically many bits of entropy compared to the uniform distribution. Dodis, Ristenpart and Vadhan (TCC ’12) constructed such seeded condensers for efficiently samplable sources that can depend arbitrarily on the seed assuming near-optimal security of collision-resistant hash functions. In this work, we investigate whether such condensers can be made seedless. We present several new results: