Unsealed Emails Reveal OpenAI and Microsoft Knew AI Was Creating a 'Doom Loop' That Would Cannibalize the Web
Unsealed internal communications in The New York Times copyright lawsuit reveal that senior executives at Microsoft and OpenAI privately acknowledged large language models risk destroying their own content supply chains. Terming web-scale scraping the 'largest theft of labor in human history,' internal assessments warned that conversational zero-click answers would trigger a systemic doom loop, bankrupting the journalists and digital publishers essential for training future frontier models.

By Ajinkya Pawar
Head of Search & AI Intelligence • The AI NEWS
Key Developments & Executive Briefing
Internal Recognition of Content Supply Chain Destruction
Unsealed Admissions'Doom Loop' WarningUnsealed filings reveal Microsoft Applied Science Director Brent Hecht and senior engineers warned that diverting publisher traffic to conversational AI answers creates a destructive feedback loop that damages model performance and the broader internet.
Documented Extraction Across Hundreds of Thousands of Works
Copyright ScrutinyProject Mango EvidenceCourt filings show OpenAI and Microsoft assembled joint datasets, including Project Mango with 160,903 publisher works and Common Crawl scrapes holding over 2 million New York Times documents, while removing copyright metadata.
End Products Devouring Their Own Essential Suppliers
Economic ParadoxLabor CannibalizationInternal Microsoft memos acknowledged the unprecedented economic dynamic where generative AI platforms disrupt the livelihood of content creators whose labor is indispensable for pre-training and real-time grounding.
For more than three years, the corporate narrative presented by frontier artificial intelligence laboratories has remained remarkably uniform: large language models are benevolent transformative engines that democratize knowledge, empower creators, and coexist harmoniously with the open web under the legal doctrine of fair use. However, unsealed internal documents filed in federal court have shattered that carefully manicured facade.
Unredacted legal exhibits made public in The New York Times copyright lawsuit against OpenAI and Microsoft reveal that leaders inside both tech giants were fully aware of the catastrophic structural consequences their technology would unleash. Rather than viewing generative AI as a symbiotic layer atop the web, internal research memos and executive exchanges frankly acknowledged that deploying conversational answer engines would create an economic 'doom loop'—systematically starving digital publishers of traffic, destroying the labor supply of professional writers, and ultimately degrading the quality of the AI models themselves.
The Anatomy of the 'Doom Loop'
At the center of the unsealed discovery materials is a January 2024 strategic assessment prepared by Brent Hecht, Microsoft's Director of Applied Science. In the briefing, Hecht warned that generative AI interfaces such as Microsoft Copilot and OpenAI ChatGPT were actively diverting traffic away from primary publishers, threatening to collapse the web's foundational business model.
Hecht termed this trajectory a 'doom loop,' explaining that as AI systems answer user questions directly without clicking through to source links, digital publishers lose the traffic necessary to sustain ad impressions and subscriber growth. Without viable revenue, newsrooms and content platforms must lay off journalists and curtail original reporting. The downstream consequence, Hecht noted with startling clarity, is that future AI models will have an impoverished data corpus from which to learn and retrieve fresh ground truth, causing model performance across the industry to deteriorate.
'It is highly unusual that an end-product threatens the economic foundations of its essential suppliers,' one internal Microsoft document stated, summarizing the paradox. 'But that is the situation we have created for our LLM business with respect to its content supply chain.'
Scraping, Paywalls, and 'The Largest Theft of Labor'
The unsealed records also expose the stark dissonance between Microsoft and OpenAI's public defenses and their private acknowledgments regarding data extraction. While corporate spokespersons have argued in court that training AI on public web pages is transformative fair use akin to a human reading books in a library, internal communications reveal employees and executives harboring severe moral and legal reservations.
According to the court filing, an internal Microsoft communication bluntly described the mass uncompensated scraping of web content as the 'largest theft of labor in human history.' Engineers and product leads observed that the commercial value captured by AI tech conglomerates was directly derived from hundreds of millions of human work-hours, while offering virtually zero attribution or financial return to the underlying creators.
Furthermore, the filings lay bare the sheer scale and deliberate mechanics of the extraction infrastructure. Evidence submitted by publishers highlights joint initiatives like 'Project Mango,' which ingested at least 160,903 unique works from news publishers, alongside custom Common Crawl slices holding over 2 million articles from The New York Times alone. Internal logs detail instances where crawlers bypassed paywalls, stripped away copyright management information, and deliberately ignored restrictive web standards to harvest high-density text corpora.
The Collapse of the Search Settlement
For nearly three decades, the relationship between search engines and web publishers operated under an implicit commercial pact: search crawlers indexed publisher content for free in exchange for delivering high-intent referral traffic. Digital media companies allowed Google and Bing to crawl their sites because those clicks translated into programmatic ad revenue, affiliate conversions, and paying subscribers.
Generative AI unilaterally broke this historic settlement. By synthesizing complete summaries and agentic workflows, conversational search engines effectively capture 100% of user dwell time while offloading none of the economic upside back to the content source. The unsealed documents confirm that OpenAI and Microsoft recognized this traffic cannibalization was not an accidental byproduct, but an inherent, anticipated characteristic of zero-click answer engines.
As the New York Times litigation proceeds toward summary judgment and potential jury trial, these unvarnished internal admissions weaken the defense that tech companies acted in good-faith reliance on fair use precedents. When an industry's own scientists and product directors put in writing that their systems cannibalize their suppliers and constitute an existential threat to the web, arguing that such behavior is merely benign reading becomes nearly impossible to defend.
The Strategic Imperative for Publishers and Search Strategists
The revelation that AI leaders anticipated the collapse of the open web demands an immediate strategic pivot from publishers, enterprise brands, and SEO practitioners. Relying solely on legacy organic search visibility is no longer a viable long-term strategy in a market governed by zero-click AI overviews and synthetic synthesis.
To survive the accelerating contraction of the open web, media organizations and brands must aggressively pivot toward closed-loop distribution channels: proprietary mobile applications, direct newsletter relationships, specialized community platforms, and cryptographic data licensing consortiums. The era of gifting open web content to algorithmic crawlers in exchange for ephemeral search traffic has definitively ended.
---\n### Fact-Checked Sources & Verified References
- The Verge - OpenAI and Microsoft knew they were starting a 'doom loop' for the web
- 404 Media - 'Doom Loop': OpenAI and Microsoft Admits LLMs Are Destroying the Web and Built on Theft
- Fast Company - Microsoft exec warned AI could create a 'doom loop' for the internet
Sources & References
Related Coverage
Discussion (0)
Be the first to share insights on this story.