OpenAI and Microsoft Knew AI Would Destroy the Web’s Future, Yet They Moved Forward

Leaked court documents suggest that **OpenAI** and **Microsoft** were aware that the development of artificial intelligence could generate a "doom loop" detrimental to both media organizations and the AI companies themselves, by reducing the need to visit original information sources.

## The AI and Web's "Doom Loop"

A series of court documents brought to light that **OpenAI** and **Microsoft** possessed internal information warning about the destructive potential of artificial intelligence for the World Wide Web itself. Apparently, within **Microsoft**, analysis had already been conducted on how the development of large language models could trigger a vicious cycle. This phenomenon, internally described as a "doom loop," simultaneously threatened online media and the companies developing these technologies. The concern centers on how AI models are trained using vast amounts of information extracted from the web, while products like **ChatGPT** or **Copilot**, along with summaries generated directly in search engines, decrease the need for users to visit the pages that produce that content. If these pages lose traffic and revenue, they lose their capacity to generate new information, thereby degrading the very data source upon which future generations of AI depend.

## Threat to the Content Supply Chain

One internal document cited in the legal proceedings acknowledged that the content strategy based on artificial intelligence was creating a structural problem. The final product—the AI itself—could end up threatening the economic viability of those providing the raw material for its construction. **Microsoft** reportedly described its own language models as a product capable of "destroying its own supply chain," given that they partially replace the content they later need for training and improvement. This scenario is directly related to the phenomenon known as "Google Zero," where users increasingly obtain answers from search engines or AI assistants without needing to visit the original pages. **Microsoft** CEO, **Satya Nadella**, reportedly acknowledged that this new behavior is changing how people search for information. Documents from **OpenAI** cited in the case note that once a chatbot provides a satisfactory answer, the user has fewer incentives to access the original link.

## Loss of Traffic and Protected Content Use

The economic and media specialists at **OpenAI** also allegedly analyzed this loss of traffic. According to the documents compiled, systems like **Google AI Overviews** could have caused reductions of up to **60%** in certain search referrals. This represents a significant problem for media outlets that rely on organic traffic to finance themselves through advertising, subscriptions, or other business models. Another delicate point in the proceedings affects how **OpenAI** and **Microsoft** obtained data to develop their models. The account from the **New York Times** includes critical internal statements, such as that of **Brent Hecht**, Director of Applied Science at **Microsoft**, who questioned both the scale of content collection and the defense that this usage automatically fell under the American concept of "fair use." **Microsoft**, however, has attempted to distance itself from these claims, stating that **Hecht** was hired precisely to provide contrasting academic perspectives and that his comments do not represent the company's official legal position. The court documents also note an internal concern within **OpenAI** regarding the capacity of its models to memorize and reproduce protected content almost literally. Company employees acknowledged internally that avoiding memorization was crucial for reducing potential copyright infringement, although the legal brief includes several examples where **ChatGPT** allegedly reproduced extensive fragments from various media publications.