Microsoft Seeks Summary Judgment in AI Copyright Lawsuit as It Launches OpenAI's GPT-6 Astra

Microsoft has filed for summary judgment in a New York federal copyright lawsuit, asserting that its Copilot AI assistant rarely duplicates copyrighted text verbatim. In a legal memorandum submitted on September 4 in Manhattan federal court, the company contended that using published books to train large language models is protected under the U.S. fair use doctrine.

According to court filings cited in reporting first published by The Next Web, an expert evaluation of 8.2 million Copilot user conversations identified only 24 instances where outputs matched passages from the disputed books. That rate equals 0.00029 percent of total interactions analyzed. For 202 of the 212 books cited in the litigation, the expert found no verbatim matches whatsoever.

The lawsuit, presided over by Judge Sidney Stein as part of a multidistrict litigation proceeding, aggregates legal claims from writers and publishing groups, including 400 local news organizations. Microsoft is asking the court to rule on the fair use question as a matter of law, seeking a definitive legal resolution without going to trial.

The legal maneuver coincided with Microsoft's commercial release of OpenAI's newest artificial intelligence model. On September 3, one day prior to its summary judgment motion, Microsoft made GPT-6 Astra available through its Azure Foundry Limited Access Program, setting pricing at $10 per million input tokens.

Built for automated computer execution, GPT-6 Astra can analyze on-screen information and perform operations within authorized application interfaces. Azure product documentation notes that the system can modify records, conduct software tests, and navigate developer tools. To mitigate risks, Microsoft implemented scoped access credentials, mandatory human approval for critical actions, and comprehensive activity logging, while confirming that user prompts and outputs are excluded from future model training.

Under OpenAI's Preparedness Framework, Astra became the first model to receive a "Critical" risk rating for cybersecurity. The designation indicates that the AI system is capable of detecting previously unidentified vulnerabilities and constructing functional exploits without requiring step-by-step human direction.

While Microsoft relies on low output duplication rates to bolster its U.S. fair use argument, European legal precedents treat model memorization differently. On July 31, the Munich Regional Court ruled in GEMA v. Suno—a decision currently under appeal—that storing copyrighted works within a model's parameters constitutes an infringement of reproduction rights at the time of ingestion, irrespective of output frequency. European copyright law contains no direct equivalent to the American fair use doctrine.

Furthermore, Article 55 of the European Union's AI Act imposes specific obligations on general-purpose AI models that carry systemic risk, keying regulatory duties to inherent capabilities rather than output statistics. Developers of covered models must perform documented adversarial testing, promptly report major security incidents to the EU AI Office, and maintain strict physical and digital security for model weights and infrastructure.

Enterprise deployment of Astra is currently limited to Global and U.S. data zones. Microsoft omitted an EU data zone option at launch, despite having offered European data zone coverage on Azure since November 2024, amid continuing debates over European cloud independence and AI governance.