ESPN tests AI tool to detect poker players' tells
ESPN's broadcast of the 2026 World Series of Poker Main Event featured an experimental AI tool designed to detect players' physical tells, highlighting the growing reach of computer vision.
During the early days of the 2026 World Series of Poker Main Event, ESPN integrated a novel AI tells detection tool into its live broadcast. Developed by Luke Geel, an artificial intelligence engineer for the United States Air Force, the system analyzed video feeds from the tournament to build a database of physical tells. By tracking visual inputs such as eye movements, blinking rates, posture, chip handling, and hand fidgeting, the software generated a real-time hand strength model to predict whether a player held a strong made hand, a drawing hand, or a bluff.
Despite its high-tech appeal, the tool faced immediate skepticism from professional players due to data constraints. The 2026 Main Event attracted more than 9,000 entries, but only a fraction of those competitors spent time at the three tables equipped with broadcast cameras. Seventeen-year poker veteran Michael Gagliano, who entered the final table in eighth chip position competing for the $10 million first-place prize, noted that the limited broadcast footage made it difficult to gather actionable intelligence on opponents. Shaun Deeb, a two-time WSOP Player of the Year who finished 15th in the tournament, added that critical physical indicators like pulse rates, breathing, and leg movements cannot be captured by standard broadcast cameras.
For AI practitioners and computer vision developers, this deployment highlights the challenges of applying machine learning to high-stakes, low-data environments where human psychology is paramount. While Omaha Productions, the company licensed by ESPN for the coverage, chose not to use the tool during the final table, the experiment points to a future where AI could analyze archival footage of elite players. In exclusive high-roller tournaments where a small pool of professionals repeatedly compete on camera, thousands of hours of footage could eventually train highly specialized models. However, because electronic devices remain strictly banned at physical tables, the immediate utility of these systems remains confined to off-table preparation and broadcast entertainment.
This is our own summary of reporting by WIRED AI
