One option for preventing undesirable interactions is to make the organism (particularly its genetic information) chemically unrecognizable to other life and vice versa. This mutual incompatibility (and ideally inactivity) is termed bio-orthogonality, and it could function as an excellent partition between GMOs and their environments.
Looking for orthogonality is like asking, “What kinds of variations are permitted in the chemical conventions of life? Can we create organisms based on unnatural nucleic acids, amino acids, enzymes, metabolites, and so on?”
Genetic bio-orthogonality might just be feasible. Last year, Craig Venter announced the first "synthetic cell," actually a demonstration of inserting a synthetic genome into an already-functioning cell. This summer, another group demonstrated a bacterium evolved in the lab to preferentially use chlorouracil in its DNA in place of thymine. Taken together, these achievements look like milestones on the way to an organism that stores its genes in such a way (unnatural nucleic acids) that no other species can make use of its information and vice versa.
But unnatural nucleic acids aren't quite that easy: Chlorouracil can also be mistakenly integrated into mammalian genomes, with mutagenic side effects. This raises several points: The orthogonal organism should not produce toxic wastes or require toxic (or expensive) compounds to grow, and it should not produce toxic degradation products if killed. However, compounds that are very similar to existing metabolites are likely to be naturally functional or poisonous inhibitors/mutagens.
Modifications to the DNA backbone—using a hexose, “locked” ribose, or threose instead of ribose, for example—have been shown to be possible, but differences in flexibility, base-pairing, and melting temperature make for a lot of potential snags at the genome level. Furthermore, the more orthogonal you get, the more enzymes you have to engineer; not only do polymerases need to be altered but also ribosomes, tRNA synthases, nucleases, and maybe even kinases and other enzymes. The chemically incompatible life form is quite a piece of work.
A more reasonable approach is to meet halfway on the orthogonality. Instead of making the nucleic acids themselves orthogonal, make the code they carry orthogonal. Switching GGN from glycine to, say, proline, and so on is entirely possible with the “synthetic cell” mentioned earlier; just change all of the coding sequences to match the altered tRNA and jump right into troubleshooting! Such a system gains most of the benefits—like viral resistance and reduced gene flow—with far less enzyme engineering.
Even so, the demand for such a platform is far outweighed by the costs of making it—for now. But as gene synthesis becomes cheaper and more of the development process becomes automated, the orthogonal organism could become a reality. As difficult as building the fence might be, making the neighbors well-behaved is even harder.