Microsoft and OpenAI executives recognized a serious problem with the way artificial intelligence systems collect and present news: the technology can take value from journalism while sending little value back. Internal documents and court materials describe a system that may weaken the news industry, reduce the information available online, and eventually hurt the AI models that depend on that information.
Brent Hecht, Microsoft’s director of applied science, gave the problem a stark name: “Our AI content strategy has started a ‘doom loop.’” In his view, AI systems scrape news and other web content, answer questions without sending readers to the original publisher, and weaken the businesses that produce new material. If that process continues, the web loses the supply of reliable information that AI models need.
The news industry loses the click
Microsoft CEO Satya Nadella testified that chatbots have already acted as substitutes for news platforms. Instead of sending a reader to an original article, a chatbot can provide the information directly inside the AI platform, “giving you the information right there on the website on the AI platform versus needing to go to the underlying source.” That change removes the click that helps news organizations earn revenue.
OpenAI executive Nick Turley described the relationship in even more direct terms. He wrote that AI products are “largely substitutive” to journalism and “will get more and more substitutive as they get better.” In another statement, he called chatbots “largely substitutive, period.” An OpenAI software engineer wrote, “no matter how prominently we show the links, users won’t click.”
The available figures show the scale of that concern. Microsoft recorded 83–93 percent drops in click-through rates for some news plaintiffs and 51–94 percent drops for others. Those losses point to a decline in traffic and news revenue, especially when a chatbot gives users an answer without requiring them to visit the publisher that created the reporting.
News organizations argue that chatbot answers containing extensive verbatim overlap with their articles should not count as fair use. Their position focuses on the way these systems can reproduce reporting while keeping readers away from the original source.
Executives described the training process as theft
Hecht called AI scraping of news “an astonishing theft of unprecedented proportions” and described it as perhaps the “largest theft of labor in human history.” A Microsoft internal document from 2023 warned that “millions of people around the world will soon consider large models ‘hoovering up’ all their work to be an astonishing theft of unprecedented proportions.”
That document also included a cartoon showing large language models destroying their own supply chains. The message was simple: AI systems depend on a steady flow of human-created material, but their methods can damage the people and organizations that produce it.
Microsoft and OpenAI obtained training content by scraping millions of documents, bypassing paywalls, and removing copyright notices from training data. Their internal documents and communications showed awareness that these practices could be considered theft of news content. They also anticipated that verbatim outputs could harm news sites and tried to make it harder for news organizations to test chatbots.
The training data questions went beyond ordinary web collection. Microsoft and OpenAI did not license news content, and they obtained datasets without proper authorization. One dataset used by OpenAI contained 1.8 million articles from a third party that had an agreement not to use the data for commercial purposes.
A warning for AI companies and publishers
The documents show that the conflict is not limited to copyright disputes. It reaches the basic business model behind online journalism. News organizations spend money and labor gathering information, while AI products can summarize or reproduce that work in a format that keeps users inside the AI platform.
That is the substitution Nadella described, and it explains why executives inside Microsoft and OpenAI focused on lost clicks. If readers stop visiting news sites, publishers lose a path to revenue. If publishers reduce reporting, the web contains less new material for future AI training. The same process that helps models answer questions today can damage their supply of information tomorrow.
OpenAI cofounder and president Greg Brockman responded “ah nice.” The internal exchanges involving Brockman, OpenAI CEO Sam Altman, Microsoft, and OpenAI show that senior figures were close to a debate over the risks created by AI content systems.
Alex Haurek, a Microsoft spokesperson, said Nadella’s comments reflected broad principles and changes in how people find and consume information, not legal conclusions. That distinction leaves the central business question open: can AI companies provide useful answers without taking the value that supports the original work?
The internal record offers an uncomfortable answer. Microsoft and OpenAI understood that their systems could replace news platforms, reduce clicks, reproduce articles, and weaken journalism. They also recognized that damaging news publishers could eventually damage the web and the AI models built on it. That is the “doom loop” Hecht described: AI takes from the information system, then risks destroying the system it needs to keep working.
Based on
- Microsoft Director Admitted in Internal Document That AI Is Creating a “Doom Loop” That Destroys the News Media by Stealing Its Content Without Providing Anything in Return, in Turn Making the AI Useless — futurism.com
- Microsoft Executive Called OpenAI’s Web Scraping The ‘Largest Theft Of Labor In Human History’ — engadget.com
- Microsoft exec called AI scraping the “largest theft of labor in human history” – Ars Technica — arstechnica.com
- Microsoft exec called AI the ‘largest theft of labor’ in history, court records show – The Washington Post — washingtonpost.com
