Korean courts launch AI pilot benches for patents

Korea’s Patent Court quietly became one of the first higher courts to operationalise generative AI inside the judicial workflow, not as a research pilot in the back office but as a formal “AI division” with documented procedures and consent requirements. That matters because courts are among the most conservative institutions in any democracy. Once they bless AI as a legitimate tool for drafting and translation, it lowers the perceived barrier for other high‑stakes domains like tax, immigration or social‑security adjudication. The design choices are instructive: judges retain full control over reasoning, parties can opt out, and AI is used for summarisation, translation and stylistic editing rather than outcome prediction.

For AGI watchers, this is an early glimpse of how advanced systems might be woven into decision‑making structures that still insist on a human face. Today it is ChatGPT, Claude and similar tools helping produce draft judgments; in a few years, more capable models could propose legal theories, spot inconsistencies across thousands of cases or simulate likely appellate reactions. That could increase throughput and consistency, but it also raises deep questions about explainability, bias and the line between assistance and delegation. Whoever supplies the most reliable legal‑domain models may gain a durable foothold in government workflows that tend to be sticky once embedded. The Korean experiment will be closely watched by other jurisdictions considering how far to let AI into the courtroom without undermining due‑process norms.