Frank wants to make sure he has this right before he says anything.
The Office of Management and Budget, which is the agency responsible for overseeing federal spending, has proposed using an AI system to review and approve federal grant applications. A congresswoman asked the OMB director directly whether this AI system would be accountable to Congress. The answer was no. She asked whether it would be auditable. The answer was essentially not in the way she meant. She asked why an AI system that is neither accountable nor auditable was an appropriate mechanism for making policy decisions about who receives federal funding.
Frank read the transcript. Frank is still thinking about the answer.
The same government that took 43 days to agree on a continuing resolution wants to delegate grant approval decisions to an AI system that Congress cannot audit and that is not accountable to the people whose tax dollars are funding the grants.
Frank has questions. Frank will share them in an orderly fashion.
Question 1: Who is accountable when the AI is wrong?
Federal grants fund research, infrastructure, community programs, veterans services, scientific studies, and dozens of other things that real Americans depend on. When a human being at a federal agency makes a wrong call on a grant application there is a process. There is a record. There is someone whose name is on the decision. There is a chain of accountability that leads somewhere.
When an AI system makes a wrong call the accountability chain leads to an algorithm. An algorithm written by someone, trained on data selected by someone, calibrated to priorities set by someone, and deployed by someone. But when the congresswoman asked who is responsible for the AI's decisions the answer got complicated very quickly.
Frank believes in accountability. Frank believes that every decision made with public money should be traceable to a human being who can be questioned about it, removed from office if necessary, and held responsible for the outcome. Frank does not believe an algorithm satisfies this requirement regardless of how efficient it is.
Question 2: What are the AI's priorities and who set them?
The congresswoman asked specifically whether the AI would flag grants as inconsistent with administration priorities. The answer suggested it might. Frank wants to understand what administration priorities means in the context of an AI system reviewing grant applications for scientific research.
The National Science Foundation has already seen a 34% decrease in new awards in 2026 compared to previous years. Frank does not know exactly what the AI is doing or not doing in that process. Frank does know that a 34% decrease in scientific research funding while an AI system reviews applications for consistency with administration priorities is a combination of facts that Frank finds worth examining.
Frank is not saying the AI is the cause. Frank is saying the question deserves a better answer than it received.
Question 3: Who audits the auditor?
Frank read the Constitution again this week. Frank does this regularly. Article I gives Congress the power of the purse. The ability to control government spending is one of the most fundamental checks in the entire system. The Founders put it in Congress specifically because they wanted the spending decisions to be made by the branch most directly accountable to the people.
An unauditable AI system reviewing grant applications is not what Article I describes. It is not what the Founders had in mind when they designed a system where every expenditure required an appropriation made by law. The law is made by Congress. The AI is not accountable to Congress. Frank finds this gap significant.
Frank's actual position on AI.
Frank is not anti-technology. Frank uses technology. Frank dictates this column on a device that would have seemed like science fiction when Frank was in high school. Frank is aware that AI tools can be genuinely useful for processing large amounts of information quickly.
What Frank is against is using technology to obscure accountability. What Frank is against is pointing at an algorithm when a constituent asks why their grant application was denied and saying the system flagged it. What Frank is against is building a layer of technical complexity between the government and the governed that makes it harder to ask who decided this and why.
The government works for the people. The people paid for the grants. The people deserve a human being they can question when the decision goes wrong.
Frank will be watching this one closely.
— Frank