AI in Permitting: The Governance Question Nobody Wants to Answer
May 11th, 2026
AI in permitting just crossed into new territory. Not because of a product launch or a funding round, but because of two governance signals that arrived in the same week.
First: GL Solutions' 2026 state-by-state assessments report that AI has replaced cybersecurity as the top priority for state government officials for the first time in 12 years. Second: the Secretary of the Interior issued a Secretarial Order that explicitly names "permitting efficiency" as a domain for AI deployment, claiming DOI is "already seeing results," including "streamlined environmental reviews."
These aren't pilot announcements. They're institutional commitments. When a Cabinet secretary issues an order naming AI in permitting and 50 state CIOs rank it above cybersecurity, the question shifts from "should we do this?" to "who's responsible when it goes wrong?"
And it will go wrong somewhere. Utah suspended an AI prescription renewal program in January, just months after deployment. The details matter: the system was making decisions in a high-stakes regulatory domain without sufficient oversight mechanisms. Utah pulled it back. That's the right response, but it happened after deployment, not before.
Beveridge & Diamond published a legal analysis this month that flagged a risk the permitting community hasn't absorbed: project opponents may use AI involvement in environmental review as grounds for legal challenge. The argument would be straightforward — if an AI tool contributed to an environmental assessment and the assessment is challenged in court, the agency will need to demonstrate that human reviewers exercised independent judgment at every decision point. If they can't, the review is vulnerable.
This isn't hypothetical. NEPA litigation is built on procedural adequacy. Courts don't review whether the agency reached the right conclusion — they review whether the agency followed the right process. An AI tool that drafts language, suggests findings, or routes decisions creates a new surface area for procedural challenges. Did the reviewer independently evaluate the AI's output? Can the record demonstrate that? If the agency can't answer those questions, the tool becomes a liability.
The DOI Secretarial Order tries to address this by requiring "human-in-the-loop safeguards." That's the right instinct, but it's not a governance framework. "Human in the loop" can mean anything from "a person clicks approve" to "a senior reviewer independently evaluates every AI-generated output against the source material." The difference between those two definitions is the difference between a defensible record and a court loss.
What's missing is the layer between "we have an AI tool" and "we have a process for using it that will survive legal challenge." That layer includes decision logs that document where AI was and wasn't used, training records showing reviewers understand the tool's limitations, standard operating procedures for when to override AI suggestions, and retention policies for AI-generated drafts and intermediate outputs.
None of that is technology. All of it is governance. And almost no agency deploying AI in permitting has built it yet.
The agencies that get this right will be the ones that treat AI governance as a precondition for deployment, not a cleanup task after the first lawsuit. The ones that don't will learn the lesson the way Utah did — publicly, expensively, and after the damage is done.
What to watch: whether any federal agency publishes AI use documentation standards for environmental review. Whether the DOI's Secretarial Order produces implementing guidance with specific procedural requirements. And whether the first NEPA challenge citing AI involvement in the review process arrives in 2026 or 2027.
GL Solutions · Beveridge & Diamond · DOI Secretarial Order
AI in Permitting is a recurring column on Permitting Tech covering how artificial intelligence is entering permitting workflows. Written by Boon Sheridan.
