The immune systems that protect organisms from infectious agents invariably have a cost for the host. In bacteria and archaea CRISPR-Cas loci can serve as adaptive immune systems that protect these microbes from infectiously transmitted DNAs. When those DNAs are borne by lytic viruses (phages), this protection can provide a considerable advantage. CRISPR-Cas immunity can also prevent cells from acquiring plasmids and free DNA bearing genes that increase their fitness. Here, we use a combination of experiments and mathematical-computer simulation models to explore this downside of CRISPR-Cas immunity and its implications for the maintenance of CRISPR-Cas loci in microbial populations. We analyzed the conjugational transfer of the staphylococcal plasmid pG0400 into Staphylococcus epidermidis RP62a recipients that bear a CRISPR-Cas locus targeting this plasmid. Contrary to what is anticipated for lytic phages, which evade CRISPR by mutations in the target region, the evasion of CRISPR immunity by plasmids occurs at the level of the host through loss of functional CRISPR-Cas immunity. The results of our experiments and models indicate that more than 10(-4) of the cells in CRISPR-Cas positive populations are defective or deleted for the CRISPR-Cas region and thereby able to receive and carry the plasmid. Most intriguingly, the loss of CRISPR function even by large deletions can have little or no fitness cost in vitro. These theoretical and experimental results can account for the considerable variation in the existence, number and function of CRISPR-Cas loci within and between bacterial species. We postulate that as a consequence of the opposing positive and negative selection for immunity, CRISPR-Cas systems are in a continuous state of flux. They are lost when they bear immunity to laterally transferred beneficial genes, re-acquired by horizontal gene transfer, and ascend in environments where phage are a major source of mortality.