Provisional Guidance for Users of LLM-Based Code Generators — /dev/lawyer

Mitchell maintains GhosTTY, a terminal emulator and supporting library. GhosTTY is MIT-licensed, but many of the older emulators and libraries it competes with are under GPLs. Mitchell is a very active user of large-language-model-based code generators like Copilot. He wishes there were some clear legal decision on whether using code generators trained on GPL-licensed code causes copyright issues, one way or another, so at least it’s clear what he should and shouldn’t do.

Mitchell is right that a comprehensive decision like that doesn’t exist yet. There are lawsuits in the courts that might give us something close, eventually. But experience suggests it will take some time, and also that clarity will likely come piecemeal, in decisions that feel incomplete and come laden with practical asterisks. That’s little consolation to devs like Mitchell, who just want to know what the rules are, so they can play their best honorable game.

I can’t usurp public authority, or pretend to perfectly foresee what those in robes will say. But here is what I would personally tell developers overall, right now, based on my own judgment and reading.

Leave a Comment