When Jeffrey Epstein died in 2019, he left behind an enormous digital archive: more than 1.4 million emails. Reading them manually would have taken years. So we decided to let artificial intelligence do the heavy lifting.
We converted the emails from scanned PDFs into searchable text and fed the entire dataset into a powerful large language model (LLM). The AI was tasked with reading every message, understanding context, and assigning each email thread an “alarm score” from 1 to 10 based on how disturbing or concerning the content appeared.
The results were eye-opening.
The majority of the 1.4 million emails were ordinary — flight arrangements, financial transfers, dinner reservations, and polite exchanges with staff. Yet the LLM consistently flagged several hundred threads as significantly more alarming than the rest. These high-scoring emails stood out because of their tone, secrecy, requests for unusual favors, and patterns of influence.
Some of the most disturbing messages revealed a carefully cultivated network of powerful figures from finance, politics, science, and technology. Epstein corresponded regularly with billionaires, academics, lawyers, and government officials, often mixing personal flattery with subtle requests or offers of access. Even after his first conviction, many continued the relationship without apparent hesitation.
The AI highlighted exchanges that showed deliberate efforts to control narratives, obscure information, or arrange private meetings with unclear motives. Redacted names and vague language appeared frequently in the highest-scoring threads, suggesting attempts to hide the true nature of certain dealings. The model also detected repeated patterns of Epstein leveraging his wealth and connections to gain favor or exert pressure.
What made these findings especially unsettling was how normalized the behavior seemed within elite circles. The emails painted a picture of a man who moved effortlessly through the highest levels of society while operating with minimal oversight or accountability.
This experiment demonstrates the growing power of large language models in investigative journalism. By processing vast amounts of data that no human team could realistically review, the AI helped surface patterns and red flags that might otherwise have remained buried.
The most disturbing revelation was not any single email, but the broader picture it created: a web of influence where money, power, and secrecy intertwined with little resistance. Jeffrey Epstein may be gone, but the systems that enabled him clearly remain intact.
Leave a Reply