The Michigan State Police demanded $6.8 million to process a Freedom of Information Act request from my organization in 2009.
A federal inspector general had flagged problems with the agency’s spending of federal homeland security grants, so we asked for documents detailing how much money the police received and how they spent it. We thought it was a simple request, but the bill was bigger than my organization’s entire budget. We dropped the request, and to this day, Michiganders are still in the dark about how a crucial part of their government acted.
I’ve been thinking about this story amid the rise of artificial intelligence. The technology is automating a huge number of tasks in the private sector, from the analysis of medical data to the creation of presentations. But one of the most beneficial uses could be to help the public sector process records requests, quickly and affordably. Artificial intelligence could usher in a new era of transparency.
WHY IS THE WHITE HOUSE TRYING TO KILL COMMON-SENSE AI LEGISLATION IN UTAH?
What happened to the Mackinac Center is far from unique. Every state has a sunshine law that requires government agencies to turn over documents when members of the press or public ask, while the federal government has the Freedom of Information Act. But records requests aren’t self-fulfilling. Assuming the government even agrees to respond, a request usually requires a painstaking search for documents by a living, breathing human being. Once the search is complete, that person, or perhaps an army of bureaucrats, must review the documents one by one, blacking out anything that can’t be disclosed.
This process is pricey, in time and money, and governments typically pass some or all the cost onto the requesting party. Last year, a Michigan township demanded $100,000 for records about the local fire chief’s suspension. In Georgia, a school district recently tried to charge a journalist nearly $3,000 to send 80,000 pages of records. Apparently, staff spent 113 hours redacting children’s and librarians’ names. Even a charge of a few hundred bucks can be enough to stifle a records request from an everyday citizen seeking the truth.
Nor does it help that more people are making records requests, creating a backlog at every level of government. David Cullier, director of the Brechner Center for the Advancement of the First Amendment, told me the federal FOIA backlog had 69,000 unfilled requests in 2010. By 2024, that number had jumped to nearly 270,000. Over the same time, the percentage of requests that were fully granted fell from 38% to 12%, while the average response time nearly doubled. The FOIA nonprofit organization MuckRock has found that at the state level, 40% of its requests were successful in 2019, but barely one-third were successful in 2025. Transparency is critical to good government, but the state of disinfecting sunshine is bad and getting worse.
Public officials should look to artificial intelligence to make things right. The technology is more than powerful enough to search public records, retrieve relevant data, redact the necessary information, and respond to public requests. Practically, governments could create online portals where anyone — yes, anyone — could make a request. That would obviously be superior to the status quo.
Where the government charges anywhere from hundreds to thousands to millions of dollars, artificial intelligence could crunch the data for pennies. Where bureaucrats take weeks or months, artificial intelligence could take seconds or minutes. Think of it like Ancestry.com or Newspapers.com, which respectively put your family history and local journalism instantly at your fingertips. Transparency could be the same. There would also be less litigation from aggrieved parties who were denied requests or faced costly demands. And government officials themselves could spend less time combing through documents with a blackout pen.
To be sure, such a system would need to be expertly crafted. Artificial intelligence could accidentally disclose sensitive personal or confidential information, affecting everything from public safety to national security. But the current process already suffers from the same errors — it’s run by fallible humans. With the proper guardrails written into the code, and with continual updates to ensure accuracy, artificial intelligence could become the safest option, to say nothing of the fastest and cheapest.
AMERICA’S AI FUTURE REQUIRES MASSIVE INFRASTRUCTURE INVESTMENT
The winds of change are beginning to blow. The National Archives and Records Administration reports that nearly one-fifth of federal agencies have started using machine learning to help with FOIA requests, but that’s a far cry from letting the public access records without the government playing gatekeeper. States and local governments have made hardly any moves in this direction, even as response rates to record requests keep sliding. For the sake of transparency and public trust in government, elected officials should make AI-driven sunshine programs an immediate priority.
The Michigan State Police would be a good place to start.
Michael J. Reitz is executive vice president of the Mackinac Center for Public Policy.