Microsoft Application Science Director Calls AI Model Training the Greatest Labor Theft in Human History
Brent Hecht, Microsoft’s Application Science Director, referred to AI model training as "the greatest labor theft in human history" in internal documents.
This statement comes from legal briefs unsealed in September related to the copyright lawsuit involving Microsoft and OpenAI, where Hecht also described large-scale content scraping as "an astonishing theft of unprecedented proportions" and warned of a potential "doomsday loop" harming models and the entire web.
The documents show he questioned the fair use argument, stating that the widespread scraping of news makes the concept of fair use a complete mockery, and pointed out that almost no one intends for their creative content to be used or compensated in this way.
Microsoft stated that these memos do not represent the company's views, while executives at OpenAI have also warned that AI poses an existential threat to publishers, with Copilot data showing click-through rates dropping by as much as 93%.
The plaintiffs in the lawsuit include publishers like The New York Times, accusing the defendants of using copyrighted content to train models without permission, while the defendants claim fair use.
The incident was driven by unsealed lawsuit documents, shifting attention from public fair use defenses to internal concerns about the disruption of the content supply chain, benefiting content creators seeking licensing and compensation, while pressuring AI companies that rely on large-scale scraping for training.
Internal discussions reveal an awareness of the threat to the content supply base posed by their own products, reinforcing the boundaries of responsibility in copyright disputes.
Source: Public Information
ABAB AI Insight
As Microsoft’s Application Science Director, Brent Hecht's historical trajectory includes directly describing large-scale content scraping as labor theft and a mockery of fair use in internal memos, foreseeing a "doomsday loop" for models, contrasting with the company's public stance on fair use.
On the capital path, the company maintains its data acquisition route through litigation defense, while internal documents show recorded negative impacts on publisher traffic and content supply, motivated by balancing model performance with supply chain sustainability, yet the public stance still advocates transformative use.
Similar cases include other AI companies facing tensions between internal documents and public defenses in copyright lawsuits, as generative AI transitions from unlicensed scraping to potential licensing and compensation negotiations.
Essentially, this involves a shift in pricing power and restructuring of the industry chain: content shifts from free input through public scraping to compensated supply, with mechanisms that directly replace supplier traffic and revenue, creating a self-undermining cycle, making AI companies responsible participants regarding their content base, and repricing copyright and labor value.
ABAB News · Cognitive Law
Training data is labor, scraping is theft
Products destroy supply, the loop closes
Internally called theft, publicly termed fair use