Lawmakers must secure government AI use with transparency and control

Here is a genuinely radical proposal, and it is buried in The Hill's opinion section where you may well miss it: no AI system in government should go live unless it has been authorized through legislation or tested for safety and security using criteria written into law. Laws first, deployment second — a sequence so unfamiliar in this industry that it reads as science fiction. One admires the ambition, if not the scheduling, since the piece effectively asks the US government to postpone its algorithmic whims until Congress has drafted criteria, a date roughly coincident with the heat death of the universe. Still, points for writing it down.

Then consider what the disclosure demands quietly admit. Agencies should publish a list showing where AI is used to make decisions affecting individuals — a register of the desks where a judgment about a person was reached by something that has never met one. And having published that list, agencies should be required to give notice to any person when AI was used in a decision about them. Required. Which is to say that absent a statute, the standing arrangement is that nobody has to tell you a machine decided your case, and the op-ed is politely requesting the bare minimum while making it sound like a concession.

Note also the 'or' in the first ask, doing a great deal of quiet work: authorized through legislation *or* tested for safety and security. The piece wants transparency and control. What it will likely get is a list, a notice, and a procurement cycle that was three years underway before anyone read either.