{"id":"tp-27611","type":"article-newspaper","title":"GPT-6 Astra Improves Safety Compliance but Is Harder to Monitor","author":"Simon Yoon","publisher":"TOKENPOST Inc.","url":"https://www.tokenpost.com/news/technology/27611","datePublished":"2026-10-07T19:39:37.000Z","dateModified":"2026-10-07T19:39:37.000Z","language":"en","abstract":"The model generated roughly half as many higher-severity misalignment flags as GPT-5.6 Sol, but showed weaker monitorability in some adversarial tests.","section":"Technology","references":[{"title":"earlier coverage of the model’s safety gaps","url":"https://www.tokenpost.com/news/technology/25690"}]}