Evolution is driven by random mutations, whose fitness outcome is tested over time. In vitro evolution of a library of a randomly mutated protein mimics this process, however, on a short time-scale, driven by a specific outcome (such as binding to a bait). Here, we used directed in vitro evolution to investigate the role of molecular chaperones in curbing promiscuity in favor of specificity of protein-protein interactions. Using yeast surface display, we generated a random library of the E. coli protein Uracil glycosylase (UNG), and selected it against various baits. Those included the purified chaperones GroEL, DnaK+DnaJ+ATP, or total protein extracts from WT or delta DnaK+DnaJ cells. We show that in-vitro evolution differs from natural evolution in cells, both physically and thermodynamically. We found that chaperones, whether purified or as part of the protein-extract, select for, and thus enrich uracil glycosylase (UNG) misfolded species during this in vitro evolution process. In a more general context, our results show that chaperones purge promiscuous misfolded clones from the system, and thereby avoiding their detrimental effects, such as forming wrong interactions with other macromolecules, including proteins, which can harm proteostasis.SignificanceMolecular recognition of proteins is fast and specific, even in the crowded milieu of the cell. This, despite the high probability of mutations to promote promiscuous binding. Therefore, without constant purge, promiscuous interactions should be the norm. Here, we show that misfolded species dominate, when a randomly mutated protein is selected for binding either purified chaperones or chaperone-containing cell extract. This suggests that owing to their ability to specifically bind misfolded protein structures, chaperones are an evolutionary force to purge mutations-destabilized, misfolded proteins and avoid non-specific interactions that may harm proteostasis. Moreover, this shows that misfolding and binding to chaperones, is the most likely outcome of random mutations, rather than the formation of new specific protein-protein interactions.