Microsoft, GitHub, and OpenAI told a San Francisco federal court to dismiss a proposed class action suit against Copilot, citing fair use of open-source code
Blake Brittain / Reuters :
Context & Ripple Effects
The lawyers behind the Copilot class action framed their theory in November as a direct challenge to how generative AI models are trained on public repositories — code scraped from GitHub without attribution or license compliance. Microsoft, GitHub, and OpenAI's answer is the industry's standard defense: training and output are fair use of open-source code.
The motion matters beyond this one case because it lands in the same docket stream as OpenAI's other copyright fights — the books case where a California unfair-competition claim survived dismissal, and the Raw Story and AlterNet suit a judge tossed for lack of shown harm — making the Copilot ruling a signal for how courts weigh training-data claims generally.
First-order effects
- The three defendants convert the developers' suit into an early fair-use test: if the San Francisco court grants dismissal, the proposed class of code authors loses its leverage before discovery ever opens.
- The plaintiffs' attribution theory — that Copilot reproduces code without license notices — now has to survive a motion arguing that both training and output fall inside fair use.
Second-order effects
- A ruling either way pressures OpenAI's parallel copyright docket: the books plaintiffs and future training-data claimants inherit whatever reasoning the court applies to code scraped at scale.
- GitHub's commercial position is on the line — Copilot is a paid product built on repository code, so an adverse ruling would force licensing or filtering changes across its developer tools.
Third-order effects
- If courts keep narrowing these suits — as later happened when a judge cut the Copilot case from 22 claims to two — the practical template for challenging AI training on copyrighted material shifts from broad copyright theories toward narrow, provable-output claims.
- The pattern pushes the industry toward either negotiated licensing markets for training data or judicially defined fair-use boundaries, with the Copilot docket serving as the reference case other plaintiffs and defendants cite.
The trend: US courts are becoming the de facto rulemakers for whether training generative AI on copyrighted material is fair use, with the Copilot class action as the code-sector test case.