One challenge in engineering organisms is taking responsibility for their behavior over many generations. Spontaneous mutations arising before or during use can impact heterologous genetic functions, disrupt system integration, or change organism phenotype. Here, we propose restructuring the genetic code itself such that point mutations in protein-coding sequences are selected against. Synthetic genetic systems so-encoded should fail more safely in response to most spontaneous mutations. We designed fail-safe codes and simulated their expected effects on the evolution of so-encoded proteins. We predict fail-safe codes supporting expression of 20 or 15 amino acids could slow protein evolution to ∼30% or 0% the rate of standard-encoded proteins, respectively. We also designed quadruplet-codon codes that should ensure all single point mutations in protein-coding sequences are selected against while maintaining expression of 20 or more amino acids. We demonstrate experimentally that a reduced set of 21 tRNAs is capable of expressing a protein encoded by only 20 sense codons, whereas a standard 64-codon encoding is not expressed. Our work suggests that biological systems using rationally depleted but otherwise natural translation systems should evolve more slowly and that such hypoevolvable organisms may be less likely to invade new niches or outcompete native populations.