Court docs reveal Microsoft exec warned AI models could hollow out the internet

Newly unsealed court documents in the New York Times lawsuit against OpenAI and Microsoft reveal internal concerns from Microsoft leadership about the long-term viability of AI training on scraped web content.

In an internal presentation, Brent Hecht, Microsoft's director of applied science, warned that Copilot's answer engine would "hurt the performance of our models and the entire web at the same time." Hecht described the dynamic as a "doom loop," where AI systems scrape written work for economic value and repackage it as easily accessible answers, undermining the original sources.

Internal Microsoft data showed the impact was significant. Click-through rates for New York Times search results dropped between 87% to 93% when using Copilot compared to standard Bing searches. For properties owned by Ziff Davis, which include Eurogamer and IGN, click-through rates fell between 51% and 94%.

In a January 2023 memo, Hecht called the practice "the largest theft of labor in human history." The filings also show OpenAI president Greg Brockman responding positively when told about "a hack to get around nytimes paywall" for scraping purposes, replying "ah nice."

Microsoft CEO Satya Nadella testified under oath that if he had known OpenAI was training on paywalled content, he would have required the company to retrain its models.