{"id":"tp-41461","type":"article-newspaper","title":"LLM Sampling Bias Raises Simulated Security Risk in Patrol Test","author":"Simon Yoon","publisher":"TOKENPOST Inc.","url":"https://www.tokenpost.com/news/technology/41461","datePublished":"2026-10-27T00:02:26.000Z","dateModified":"2026-10-27T00:02:26.000Z","language":"en","abstract":"Three Gemini variants produced modeled attacker success rates of 96.6% to 99.9%, compared with 75% under uniform server selection.","section":"Technology","references":[]}