OpenAI and PNNL Launch Benchmark for AI in Federal Permitting

OpenAI and Pacific Northwest National Laboratory (PNNL) have introduced DraftNEPABench, a new benchmark that measures how well AI coding agents can accelerate federal permitting under the National Environmental Policy Act (NEPA).

The benchmark's headline result: AI agents showed potential to reduce NEPA drafting time by up to 15%, according to the announcement from the two organizations.

The stakes are considerable. NEPA, signed into law in 1970, requires federal agencies to assess the environmental impacts of major infrastructure projects before approving them. Reviews under the statute are notoriously slow and are frequently cited as a bottleneck for roads, transmission lines, energy projects and other infrastructure. Permitting reform has become a priority for both parties in Washington, and AI-assisted drafting is one of the tools policymakers and agencies are now testing.

DraftNEPABench approaches the problem from a technical angle. Rather than evaluating a chatbot answering questions, the benchmark scores AI coding agents — systems that can read documents, retrieve information and produce structured draft text — on realistic permitting tasks. This places it in the growing family of agentic benchmarks that test models on multi-step work rather than single-turn answers.

The collaboration pairs a national laboratory with a leading AI developer, a model for how federal research institutions are working directly with commercial AI companies. PNNL, a Department of Energy laboratory operated by Battelle, brings domain expertise in energy and environmental analysis; OpenAI brings the models. The 15% figure gives agencies and researchers a concrete, measured starting point rather than a vendor claim.

The release also signals a broader push to modernize infrastructure reviews with software. If AI agents can reliably compress drafting work on environmental review documents, follow-on effects could reach queue backlogs at agencies that currently process NEPA reviews with large manual workforces.

For now, the 15% reduction is a ceiling demonstrated in a benchmark setting, not a deployed outcome. The next question is whether agencies adopt tools validated on DraftNEPABench and whether measured gains hold on live permitting dockets.