Microsoft exec called AI scraping 'the largest theft of labor in human history'
Unsealed Documents: Microsoft Executive Labels AI Scraping as "Human History's Largest Theft of Labor"
Recent unredacted court filings from the ongoing legal battle between The New York Times and the duo of OpenAI and Microsoft have brought shocking internal admissions to light. The lawsuit, now three years old, suggests that the very companies building these AI tools viewed their data acquisition methods as theft and recognized the "existential threat" their products pose to the journalism industry.
Satya Nadella during OpenAI DevDay 2023.
The "Theft" of Intellectual Property
Internal communications reveal a stark contrast between public corporate narratives and private admissions. Brent Hecht, Microsoft’s director of Applied Science, didn't mince words in a January 2023 memo, describing the scraping process as:
"An astonishing theft of unprecedented proportions" and "the largest theft of labor in human history."
The filings allege that the companies didn't just scrape public data; they actively:
- Bypassed paywalls without detection.
- Conducted mass scraping to build training sets.
- Intentionally removed copyright notices from the data.
The Economic "Doom Loop"
A central point of contention is the concept of Fair Use. While AI firms argue that training is transformative, Microsoft's own data suggests a parasitic relationship. The "answer engine" capabilities of Copilot have drastically reduced the need for users to visit original sources.
The impact on The New York Times is quantified by a massive drop in traffic:
This phenomenon was described by Hecht in a 2024 presentation as a "doom loop." He noted that it is "highly unusual" for a product to destroy the economic viability of its own "content supply chain," yet that is exactly what has occurred.
The Data Exchange Pipeline
The relationship between the two giants involved a bidirectional flow of massive datasets:
Legal Admissions and Contradictions
The unsealed documents provide evidence that may undermine the "fair use" defense, which typically requires that the new work does not substitute for the original in the marketplace.
| Executive | Key Admission | Impact on Legal Strategy |
|---|---|---|
| Nick Turley (OpenAI) | Products are "largely substitutive" and an "existential threat." | Challenges the "transformative" claim. |
| Greg Brockman (OpenAI) | Models are "excellent at news." | Suggests direct competition with news outlets. |
| Satya Nadella (Microsoft) | Paywalled content should be licensed. | Admits a standard for payment exists. |
In his deposition, CEO Satya Nadella stated that if he had known OpenAI was training on paywalled data, he would have exercised Microsoft's right to demand that OpenAI retrain its models.
The Scale of the Operation
The sheer volume of copyrighted material ingested is staggering. The filings detail the following counts:
- Mid-training datasets: Over
91,692copies of works from the NYT, Daily News, and Center for Investigative Reporting. - Common Crawl data: More than
2,000,000documents sourced specifically fromnytimes.com. - Project Mango: A specific initiative that assembled at least
160,903unique news works.
Technical Representation of Data Ingestion
If one were to represent the alleged stripping of copyright notices in a simplified logic flow, it might look like this:
def process_training_data(raw_content):
# Alleged process of removing legal markers
cleaned_data = raw_content.replace("Copyright ©", "")
cleaned_data = cleaned_data.replace("All rights reserved", "")
return cleaned_data
Conclusion: A Threat to the Creators
Beyond the financial loss, Microsoft documents admit a "real risk" that generative AI will "significantly disrupt the employment" of the very writers and journalists whose labor provided the foundation for the models. While the Trump administration has recently filed a brief supporting OpenAI's unlicensed use of data, these internal admissions paint a picture of a technology that knows it is cannibalizing its own sources.
Related Assets:
Reported by Rebecca Bellan 