{"id":"tp-29053","type":"article-newspaper","title":"AI Guardrail Tests Show Typed Decision Models Can Fail Open","author":"Simon Yoon","publisher":"TOKENPOST Inc.","url":"https://www.tokenpost.com/news/technology/29053","datePublished":"2026-10-09T14:14:24.000Z","dateModified":"2026-10-09T14:14:24.000Z","language":"en","abstract":"Seven open-weight models reached 36%–72% accuracy, while misleading option labels drove fail-open rates as high as 100% in four models.","section":"Technology","references":[]}