Microsoft Says Copilot Rarely Reproduced NYT Articles

What changed

Based on reporting by The Verge, Microsoft filed analysis from 8.2 million Copilot chat logs as it seeks summary judgment in the consolidated copyright case brought by news publishers and book authors, including The New York Times. Microsoft says 59,545 logs shared at least 16 words with news content, while an authors’ expert found 24 responses with at least 30 matching words; the Times rejects Microsoft’s conclusion that this shows Copilot does not substitute for its journalism.

Why This Matters

This publication’s view: this is less a debate about whether a chatbot can quote than a test of what counts as product competition. Microsoft is arguing that a system can train on copyrighted work and still be meaningfully different from the work itself, particularly when its outputs rarely reproduce long passages.

That distinction matters wherever AI features turn source material into answers. A low rate of matching text may be useful evidence for Microsoft, but it does not by itself settle whether a product diverts attention or value from the original publisher. The court will have to decide whether “not copying much text” is enough when the complaint is also about substitution.

Impact assessment

Microsoft and OpenAI gain a concrete exhibit for their fair-use and summary-judgment arguments: Microsoft says Copilot’s logged outputs rarely contained substantial overlap with the plaintiffs’ work.

The New York Times faces an argument aimed directly at its substitution claim. Its counsel says discovery instead shows Microsoft and OpenAI used Times material to build competing commercial products.

The Center for Investigative Reporting sits in the awkward middle: Microsoft says an expert identified 51 substantial-overlap instances involving CIR work, while CIR had not responded to The Verge’s request for comment.

Scenarios

Most likely: If the judge treats neither side’s reading of the discovery record as decisive on its own, the court considers Microsoft’s log analysis alongside the publishers’ and authors’ substitution arguments when deciding summary judgment.

Upside: If the judge accepts Microsoft and OpenAI’s fair-use argument and finds the cited overlap evidence insufficient to sustain the claims, the companies could win an early end to the case through summary judgment.

Downside: If the judge finds disputed facts about copying or substitution that require further resolution, summary judgment is denied and the consolidated litigation continues.

What to watch next

• A ruling on Microsoft’s summary-judgment request, including whether the judge relies on transformative purpose or limited output overlap.
• Responsive filings from the Times, other publishers, or book authors that challenge or contextualize Microsoft’s 8.2-million-log analysis.