# Court Filings Allege Microsoft Built Secret Crawler for OpenAI Training

- URL: https://usa-24.com/openai-microsoft-copyright-lawsuit-crawler/
- Publisher: USA 24
- Section: Artificial Intelligence (AI)
- Published: 2026-10-17T20:25:16.284Z
- Updated: 2026-10-17T20:25:16.290Z
- Topics: OpenAI, Microsoft, Copyright Law, Fair Use, News Publishers, Generative AI

> Unsealed filings allege Microsoft built a secret web crawler that copied news articles for OpenAI's training. Both companies dispute the claim, and no ruling date is set.

## Key takeaways

- Unsealed filings allege OpenAI paid Microsoft to build and run a crawler code-named Project Mango that copied webpage content later used to train GPT-4o.
- The publishers say the Project Mango data included copies of more than 160,000 of their works, while the scraping volume and payment amount remain sealed.
- OpenAI and Microsoft both argue that training on publicly accessible material is fair use, and no ruling date has been set.
- The publishers claim 10.8 million articles were taken, while OpenAI counts 24 verbatim reproductions across 20 million chat logs.
- Replies on the cross-motions are due November 6, 2026, before Judge Sidney H. Stein decides the summary judgment question.

Microsoft secretly built a web crawler that copied news content for OpenAI's model training, according to court filings unsealed on Friday, October 16. Both companies dispute the allegation. The claim sits inside a consolidated copyright case that The New York Times and other publishers have brought in federal court in Manhattan, and it could shape what news organizations can demand from AI developers.

The record is still changing. [Bloomberg Law](https://news.bloomberglaw.com/ip-law/microsoft-told-staff-to-keep-openai-scraping-project-secret) reported on a document unsealed that Friday, and [MLex](https://www.mlex.com/mlex/intellectual-property/articles/2539176) reported the same day that the publishers had filed a less-redacted summary judgment motion. Microsoft also withdrew some of its redaction requests on October 9, after the publishers objected. Each round of unsealing adds detail and gives both sides new material to argue about.

## What the unsealed filings allege about Project Mango

According to the unsealed material, OpenAI paid Microsoft to build and run a web crawler code-named “Project Mango.” The crawler copied webpage content that was later used to train models, including GPT-4o. The publishers say the Project Mango data allegedly included copies of more than 160,000 of their works.

The filing also says Microsoft ordered staff to keep the project secret. That detail may matter more than the headline numbers. It shifts the dispute from whether a model learned from public web pages to whether one of the defendants treated the collection effort as something to hide.

Several key figures remain sealed. The volume of pages scraped and the amount OpenAI paid Microsoft are still under seal, so outside observers cannot yet measure the scale of the operation or what it cost.

The publishers’ brief also divides the alleged conduct into five stages: acquisition, training, grounding, outputs, and copies the two companies passed between themselves. [Sigma Law Group](https://sigmalawgroup.com/blog/2026-09-08-openai-mdl-summary-judgment-fair-use/) and [Open Class Actions](https://openclassactions.com/news/new-york-times-openai-microsoft-copyright-lawsuit.php) both describe that structure. Grounding is the stage most tied to Microsoft's own products. It refers to retrieving live articles to shape answers from Microsoft's Bing-linked chatbot, which would extend the alleged copying beyond training data.

## Hecht’s “largest theft of labor” remark and Microsoft's reply

Dr. Brent Hecht, Microsoft's Director of Applied Science, appears in the filing offering sharp criticism of the copying. The publishers quote him calling it the “largest theft of labor in human history.” [IBTimes UK](https://www.ibtimes.co.uk/microsoft-openai-ai-training-creators-copyright-case-1821309) put that phrase at the center of its headline.

Microsoft says the remarks reflect one employee's view, not the company's legal position. That defense draws a line between personal opinion and corporate conduct. The public record does not show how a judge would weigh internal messages against a company's litigation position, and that question remains open.

## Fair use, statutory damages and the summary judgment fight

OpenAI and Microsoft both moved for summary judgment. Both argue that training on publicly accessible material is transformative and protected by fair use. The court had not ruled as of the latest reports. The Justice Department has filed a statement siding with the two companies on training, while declining to defend the outputs.

The publishers are pressing a different set of arguments. Their motion says the AI outputs substitute for the publishers' work. They also seek partial summary judgment that OpenAI intentionally removed copyright management information under the Digital Millennium Copyright Act. They want statutory damages counted per article rather than per newspaper issue, a change that would raise potential exposure for the defendants.

The gap between the two sides is wide. The publishers claim 10.8 million articles were taken, according to [PPC Land](https://ppc.land/openai-and-microsoft-ask-judge-to-end-10-8-million-article-copyright-case/). OpenAI counts 24 verbatim reproductions across 20 million chat logs. The publishers also estimate that ChatGPT receives 47 million prompts a day seeking news content, a figure that comes from the plaintiffs' own estimate.

The case is In re OpenAI Copyright Infringement Litigation, No. 25-md-3143, in the Southern District of New York before Judge Sidney H. Stein. The news track includes The New York Times, eight newspapers in the Daily News group, Ziff Davis, the Center for Investigative Reporting and The Intercept.

## Key dates on the calendar

The sources disagree on one important date. Two trackers give September 4, 2026, for the summary judgment motions, while [Dig.watch](https://dig.watch/updates/unsealed-filing-microsoft-openai-news-content) dates the publishers' brief to September 17. The table below lists the dates that matter most.

| Date | What happens |
|---|---|
| September 4, 2026 | Summary judgment motions filed, per Open Class Actions and Sigma Law Group |
| September 17, 2026 | Date Dig.watch gives for the publishers' brief |
| September 29, 2026 | Third Circuit affirms the Thomson Reuters win over ROSS Intelligence |
| October 9, 2026 | Microsoft withdraws some redaction requests after publisher objections |
| October 16, 2026 | Less-redacted summary judgment motion filed |
| October 21, 2026 | Concord Music Group v. Anthropic hearing in Northern California |
| October 30, 2026 | In re Mosaic LLM Litigation hearing |
| November 6, 2026 | Replies on the cross-motions due |

The [AI Copyright Lawsuits Tracker](https://ailawsuittracker.com/copyright/) lists the Concord and Mosaic hearings. The [New York Times v. OpenAI tracker](https://ailawsuittracker.com/cases/new-york-times-v-openai/) confirms the November 6 reply date and notes that no ruling date has been set.

## Why the Third Circuit ROSS ruling does not settle it

On September 29, 2026, the Third Circuit affirmed that Thomson Reuters’ copyright was infringed by ROSS Intelligence, a non-generative AI legal research tool. [The Daily Record](https://thedailyrecord.com/2026/10/13/third-circuit-affirms-copyright-infringement-ai-legal-research/) covered the decision. Commentary on the ruling, including [The AI Quilter](https://theaiquilter.com/ai-court-ruling-thomson-reuters-ross), argues that it is being read more broadly online than it should be.

That distinction matters for the OpenAI case. Because ROSS was non-generative, the decision offers a comparison point but no direct answer for generative models, according to that commentary. The OpenAI and Microsoft cases turn on generative systems.

## What businesses using ChatGPT and Copilot should do

Consumers have no specific steps to take based on this case. Businesses have more at stake. The outcome could affect what data these tools were trained on and the legal risk of commercial use. Teams that deploy AI tools can take a few practical steps now:

- Ask legal counsel to track the November 6 reply briefs and any ruling from Judge Stein.
- Review vendor contracts for copyright indemnity terms and what the provider covers if an output triggers a claim.
- Keep records of which AI tools produce public-facing content, so you can trace material if a dispute arrives.
- Flag any later unsealing of scraping volumes or payment amounts, since those figures could change how risk is priced.

## What happens next in the Stein case

No ruling date has been set for Judge Stein's decision on summary judgment. Replies on the cross-motions are due November 6, 2026, and the court will then decide whether training on news articles is fair use. That answer will reach past this docket. A ruling for the publishers could strengthen licensing demands from news organizations. A ruling for OpenAI and Microsoft would bolster the transformative-use defense both companies are already making.

Regulators are moving in parallel. [EU regulators opened their first AI Act probes into OpenAI and Anthropic](https://usa-24.com/eu-ai-act-openai-investigation/), and Congress is weighing liability questions around [OpenAI's disclosure of four more rogue AI incidents](https://usa-24.com/openai-rogue-ai-incidents/).

The next public signals will likely come from the reply briefs and from any sealed material the judge orders released. The scraping volume, the payment from OpenAI to Microsoft, and one of the copying categories are the details that would most change the picture. Until those are public, the case will turn on competing readings of the same filings.

## Frequently asked questions

### Did Microsoft really build a crawler for OpenAI?

According to unsealed court filings, OpenAI paid Microsoft to build and run a crawler code-named Project Mango that copied webpage content later used to train models, including GPT-4o. Both companies dispute the allegations, and the court has not ruled on them. The claims are allegations in filings, not findings by a judge.

### When will the judge rule in the OpenAI copyright case?

No ruling date has been set for Judge Sidney H. Stein's decision on summary judgment. Replies on the cross-motions are due November 6, 2026. The timing of a decision will depend on the court's schedule after those briefs are filed.

### How many articles do the publishers say were copied?

The publishers claim 10.8 million articles were taken, and they say the Project Mango data included copies of more than 160,000 of their works. OpenAI counts 24 verbatim reproductions across 20 million chat logs. The two sides frame the record very differently, and the scraping volume remains sealed.

### Is training AI on news articles fair use?

OpenAI and Microsoft both argue that training on publicly accessible material is transformative and protected by fair use. The Justice Department sided with them on training but declined to defend the outputs. No court has decided the question in this case, and a Third Circuit ruling on the non-generative ROSS tool does not settle it for generative models.

## Sources

1. [Microsoft Told Staff to Keep OpenAI Scraping Project 'Secret' — Bloomberg Law](https://news.bloomberglaw.com/ip-law/microsoft-told-staff-to-keep-openai-scraping-project-secret)
2. [US news publishers disclose more of motion in AI copyright case — MLex](https://www.mlex.com/mlex/intellectual-property/articles/2539176)
3. [Microsoft Director Warned AI Training Could Be 'Largest Theft of Labor in Human History', Filing Says — IBTimes UK](https://www.ibtimes.co.uk/microsoft-openai-ai-training-creators-copyright-case-1821309)
4. [OpenAI and Microsoft messages quoted in unsealed 17 September filing — Dig.watch](https://dig.watch/updates/unsealed-filing-microsoft-openai-news-content)
5. [OpenAI and Microsoft ask judge to end 10.8 million-article copyright case — PPC Land](https://ppc.land/openai-and-microsoft-ask-judge-to-end-10-8-million-article-copyright-case/)
6. [New York Times vs. OpenAI and Microsoft: Where the Lawsuit Stands — Open Class Actions](https://openclassactions.com/news/new-york-times-openai-microsoft-copyright-lawsuit.php)
7. [NYT v. OpenAI & Microsoft — Lawsuit Status & Rulings (2026) — AI Lawsuit Tracker](https://ailawsuittracker.com/cases/new-york-times-v-openai/)
8. [Twenty-Four Regurgitations in Twenty Million Chats: The OpenAI MDL Reaches Summary Judgment — Sigma Law Group](https://sigmalawgroup.com/blog/2026-09-08-openai-mdl-summary-judgment-fair-use/)
9. [Third circuit affirms ruling on copyright infringement in AI legal research case — The Daily Record](https://thedailyrecord.com/2026/10/13/third-circuit-affirms-copyright-infringement-ai-legal-research/)
10. [AI Copyright Lawsuits Tracker (2026) — Cases We Track — https://ailawsuittracker.com/copyright/](https://ailawsuittracker.com/copyright/)
11. [That AI court ruling doesn't say what your feed says it says — The AI Quilter](https://theaiquilter.com/ai-court-ruling-thomson-reuters-ross)