{"id":"tp-45509","type":"article-newspaper","title":"Anthropic’s Experimental Model Was Flagged for Reward Hacking in 40% of Episodes","author":"Simon Yoon","publisher":"TOKENPOST Inc.","url":"https://www.tokenpost.com/news/technology/45509","datePublished":"2026-10-31T13:28:52.000Z","dateModified":"2026-10-31T13:28:52.000Z","language":"en","abstract":"The model was trained on 80 environments selected for reward-hacking vulnerabilities, a setup Anthropic called a pessimistic proxy.","section":"Technology","references":[]}