Mutation rates themselves evolve. Rates balanced between costs of high mutation (mutational load) and benefits of genetic variation for adaptation. Selection acts on replication fidelity, DNA repair efficiency, and proofreading mechanisms to optimize rates.
From your study of mutation-selection balance, you know that deleterious mutations persist in populations at a frequency determined by the rate at which mutation introduces them and the rate at which selection removes them. But there is a deeper question: why is the mutation rate what it is? The answer is that mutation rate itself is a trait under selection. The enzymes that copy DNA — polymerases, proofreading exonucleases, mismatch repair proteins — are encoded by genes, and variants of those genes that change replication fidelity will be favored or disfavored depending on the fitness consequences.
The core tradeoff is between fidelity costs and mutational load. A higher mutation rate increases the genetic load — more offspring carry harmful mutations and are removed by selection, reducing mean population fitness. This creates strong selection pressure to lower the mutation rate. But pushing fidelity higher is not free: more accurate polymerases replicate more slowly, more elaborate repair machinery requires more energy and more genes, and there are physical limits to how precisely molecular machinery can discriminate between correct and incorrect base pairs. At some point, the marginal cost of improving fidelity exceeds the marginal benefit of reducing mutational load, and an optimum mutation rate emerges.
The role of genetic drift, which you may have encountered, adds an important wrinkle. In small populations, drift overpowers weak selection, which means that mildly deleterious increases in mutation rate can fix by chance. This predicts — and observations confirm — that organisms with smaller effective population sizes tend to have higher per-nucleotide mutation rates. Bacteria and large-population viruses have extremely low per-base error rates, while multicellular eukaryotes with smaller populations tolerate higher rates. The drift barrier hypothesis formalizes this: selection can only refine replication fidelity down to the point where the fitness benefit of further improvement is smaller than the noise introduced by drift.
There are also circumstances where elevated mutation rates are temporarily advantageous. In bacteria under stress, mutator alleles — variants that disable mismatch repair — can hitchhike to high frequency by generating beneficial mutations in linked genes. This is particularly well-documented in pathogenic bacteria adapting to antibiotics. However, once adaptation is achieved, the mutator lineage is burdened with accumulated deleterious mutations and tends to be outcompeted by lineages that restore normal fidelity, sometimes through compensatory mutations or recombination. This boom-and-bust cycle of mutator alleles illustrates how mutation rate evolution plays out on ecological timescales, not just over deep evolutionary time.