NPR finds AI misuse drives rising sanctions in US courts

The NPR piece is a snapshot of what happens when "good enough" generative tools are dropped into a system that was built on adversarial rigor. Lawyers using chatbots to draft briefs are discovering that hallucinated citations are not an abstract failure mode—they are career‑threatening mistakes. The fact that a researcher can now count over 1,200 sanctioning decisions worldwide, with U.S. courts responsible for the majority, shows that AI misuse has moved from edge case to structural problem.

For the AGI race this is a warning shot. As models grow more capable, institutions will either learn to wrap them in procedure—verification, disclosure, audit—or they’ll react with blanket hostility. Courts are among the most conservative institutions we have; if they can’t adapt to relatively brittle GPT‑era tools, they are unlikely to tolerate more autonomous systems making or drafting decisions without strong guardrails.

The medium‑term outcome is probably not “no AI in law” but a bifurcation: commodity tools banned from formal filings, and a smaller set of tightly controlled, provenance‑aware systems that can document every source they use. That dynamic is likely to echo in other high‑stakes domains, shaping demand for more interpretable and tool‑augmented models rather than raw black‑box intelligence.