There's a quiet corner of democracy that most people never think about: public comments on federal regulations. Every time a government agency proposes a new rule — on air quality, financial disclosures, healthcare policy — there's an open window where anyone can submit feedback. It all lives on regulations.gov, and the volume is staggering. Thousands of comments per rule, many with PDF attachments running dozens of pages.

Nobody reads all of them. Not really. Agencies do their best, but when you're staring down 10,000 submissions on a single docket, the loudest and most organized voices tend to dominate. The person who wrote a thoughtful two-page letter about how a housing regulation affects their neighborhood gets drowned out by a form-letter campaign.

This is exactly where AI could do something genuinely useful — not generating comments, but reading them. Clustering themes, surfacing outliers, pulling out the substantive arguments buried in legalese. Carnegie Endowment called this "unlocking public wisdom", and Fort Collins, Colorado already used AI to process over 4,000 long-form responses on a land-use issue, preserving people's actual language rather than just counting votes.

But there's a darker side. AI-generated comments are flooding the system. Astroturfing campaigns that used to require coordination now just require a prompt. Southern California's air pollution authority reportedly scrapped a regulatory plan after receiving tens of thousands of emails that may have been AI-generated. The same tool that could amplify real voices is being used to manufacture fake ones.

I find this tension genuinely interesting. The problem isn't the technology — it's the asymmetry. Right now, the groups with resources to deploy AI at scale are using it to flood the zone, while the people who'd benefit most from AI-assisted analysis don't even know it's possible. ICF has been working on using generative AI to help agencies process comments faster, and the OECD has been tracking AI's role in civic participation globally.

The interesting question isn't whether AI belongs in the regulatory process. It's already there. The question is who gets to use it, and for what.