When The New York Times team took home this year’s Pulitzer Prize for Investigative Reporting, their deep dive into SEC crypto lawsuits leaned on more than old-school reporting. Large language models worked alongside journalists, combing through mountains of documents. AI is now part of the toolkit at the top levels of investigative journalism.
Academic research highlights that transparency in newsroom AI use remains inconsistent, with some organizations openly disclosing AI involvement and others keeping it opaque.
On the 15th floor of The New York Times Building, 150 journalists, researchers, and technologists recently gathered to talk through this new AI-powered reality. The event, co-hosted with Hacks/Hackers, took place just steps from a hallway lined with more than 140 Pulitzer portraits. The setting underscored what’s at stake as newsrooms overhaul their methods for the AI age.
AI engines tackle document overload
The Epstein Files put newsroom AI to the test. Some reports mention a release of over 3.5 million pages on January 30, 2026, but Reuters confirmed that the documents in the Epstein case were printed and bound into 3,437 volumes weighing about 17,000 pounds. That shows the scale, though Reuters did not independently verify the exact page count or release date. The U.S. Department of Justice started publishing Epstein investigation materials in late 2025. Epstein died in federal custody in 2019 while awaiting trial on trafficking charges. The official cause of death was suicide.
Newsrooms didn’t sit back. NPR’s team built an internal search tool that sorted documents by type and location, pulling out names so a search for “Prince Andrew” would also find “the Duke of York.” The Times’ AI Initiatives group rolled out the Epstein Files Engine, an internal chatbot built on LibreChat. Reporters could fire off thousands of targeted questions. More than 100 Times journalists used the engine, logging over 5,000 queries and producing at least 20 published stories.
Reuters notes that the released Epstein files mention high-profile individuals such as Donald Trump, Bill Clinton, former Prince Andrew Mountbatten-Windsor, and Elon Musk. However, being named in correspondence or flight logs does not constitute a criminal accusation.
AI is also changing how reporters manage sources. J.D. Capelouto, technology reporter and AI lead at Semafor, built a Google Apps Script to keep his source spreadsheet up to date. When he tags new contacts in his inbox, the script fills a Google Sheet with names, titles, and loglines. A sidebar assistant suggests sources for new stories, and a daily script checks news headlines against the spreadsheet to recommend topics and contacts. Capelouto admits not every suggestion pans out, but the tool has nudged him to reconnect with old sources and land fresh scoops.
These automations come with risks. Handling sensitive or anonymous sources in automated systems raises security flags. Capelouto pointed out that keeping everything inside Google’s ecosystem helps limit new vulnerabilities.
AI personas go undercover
Sometimes, AI isn’t just a backend tool-it’s a shield for journalists. Reporters at The Markup, investigating Match-owned dating apps, needed to go undercover to test moderation. Instead of risking staff or using stock photos that could harm real people, they turned to AI-generated headshots and personas. These synthetic identities fooled moderation systems and let the team create 50 different accounts for their probe.
The results were eye-opening. Even after being reported and removed, these AI personas could rejoin the platforms without changing their details. The 18-month investigation won a SABEW Award for best technology business reporting. Sisi Wei, chief impact officer at The Markup and CalMatters, said AI let the team protect staff while still running a rigorous undercover operation.
Newsroom leaders aren’t blind to AI’s limits. Afrooz Mosallaei of the Center for News, Technology & Innovation said that while AI is great for generating leads, the core work-verification, source development, editorial judgment-still needs a human touch. Hallucination bugs, “vibe-coding” mistakes, and rising token costs are real headaches. But the gains are hard to ignore: AI now marks the line between drowning in data and breaking the story first.
As newsrooms scramble to adapt, those who blend AI’s brute-force power with human judgment and ethics will pull ahead. The Pulitzer hallway at the Times may soon need more space for teams who crack this new hybrid model. AI isn’t a shortcut-it’s a force multiplier for those willing to build, test, and own their workflows. As reported earlier, even small agencies are using AI to punch above their weight. In investigative journalism, both the risks and the rewards keep growing.
