Security teams already know what happens when machines hunt for flaws in software. Now two legal scholars ask what happens when the same trick targets legislation. Felix Ringe, a visiting fellow at Stanford Law School’s liftlab, and Linus Folkerts argue that large language models can now spot inconsistencies in statutes cheaply and at scale. Courts, they warn, may not keep up with the volume.
The idea flips an old slogan. Lawrence Lessig made “code is law” famous in 1999, arguing that software shapes behavior like a legal rule. Ringe turns it around and says law is code. Both contain odd formatting rules and ritual. The comparison breaks down, however, when you picture a court as a runtime, since a trial’s outcome can vary from one run to the next, and no production system could tolerate that.
To show their point, the authors used LLMs to find possible gaps in German law, and they shared only examples they consider harmless. One concerns an amendment to the Occupational Diseases Regulation, which must cite the authority behind it. They argue it leaves out a required reference, which could weaken a confidentiality rule. In theory, a journalist could then request an unpublished statement sent to the occupational-diseases advisory council. A second example involves a transit visa rule that may be invalid for similar reasons. These remain arguments, not court rulings.
Outside lawyers give a mixed review. Wendy Musell, a San Francisco employment attorney, notes that AI is already entering the legal system, with some California courts testing it as a law clerk. She has also seen AI tools show bias and misread legal issues. Eric Goldman of Santa Clara University School of Law finds the method logical but says people must check each supposed gap. Many gaps have no real consequence, he adds, and legislatures usually fix serious ones once someone points them out.
On the researchers’ proposed remedy, Goldman is cautious. Lawmakers rarely rank gap-free drafting high, so using AI during drafting looks like a long shot in the short term. Legislative counsel, who clean up bills, could plausibly adopt it first.
What I take from this is simple. Finding a gap is cheap now, but proving it matters is still slow, expensive human work.
