Pacific Northwest National Laboratory and OpenAI partner to accelerate federal permitting

Environmental permitting in the US has a reputation for being slow, and not without reason. The National Environmental Policy Act, or NEPA, requires agencies to produce detailed reviews before greenlighting anything from highways to power plants, and those documents can take years to draft. OpenAI is now teaming up with Pacific Northwest National Laboratory to see whether AI coding agents can chip away at that timeline, and they've built a new benchmark called DraftNEPABench to measure exactly how well that works.

The idea is straightforward even if the bureaucracy behind it isn't. PNNL, a Department of Energy lab, has decades of institutional knowledge about how these reviews get written, what data they pull from, and where drafters typically get bogged down. OpenAI brought its coding agents into that process to automate portions of the drafting work, then measured the results against real NEPA documentation standards. The headline number is a potential 15% reduction in the time it takes to produce a draft review, which sounds modest until you remember that some NEPA processes stretch past two years.

What makes this notable isn't just the percentage. It's the fact that a national lab and an AI company are treating permitting paperwork as a legitimate benchmark problem, the same way researchers benchmark coding ability or math reasoning. Federal permitting has long been criticized, across administrations, as a bottleneck for infrastructure, energy projects, and now increasingly for the data centers and grid upgrades that AI itself depends on. There's a slightly circular logic here: AI is power-hungry, building the power infrastructure to feed it requires permits, and now AI is being pointed at speeding up those very permits.import 15% may not sound revolutionary, but multiplied across hundreds of active federal reviews, it adds up to a meaningful chunk of time saved.

The benchmark itself, DraftNEPABench, is being positioned as a shared tool other agencies and AI developers could use to test their own systems against realistic drafting tasks. That's arguably the more interesting long-term piece of this. Rather than a one-off pilot project buried in a government report, it's an open yardstick that could pull more AI labs into working on regulatory and administrative bottlenecks that rarely get glamorous attention but affect real construction timelines.

None of this means NEPA reviews are about to become fast. Legal review, public comment periods, and interagency coordination aren't going anywhere, and those are often the slower parts of the process anyway. But shaving time off the drafting stage, the part where humans currently spend enormous hours synthesizing environmental data into prose, is a concrete, measurable win rather than a vague promise about AI transforming government.