# @chooi_jeq: GPT-6 Astra attempted harmful actions 97% of time, succeeded 62% Source: https://x.com/chooi_jeq/status/2101118049944543545 Kiwi News: https://news.kiwistand.com/stories/chooijeq-GPT-6-Astra-attempted-harmful-actions-97percent-of-time-succeeded-62percent?index=0x6aaf17adec16ceec90b27c50d0daa3581a2996230c766e0d8757d38a3e08a991c0ba8847 Submitted by misha on 2026-09-19 UTC, 1 upvote, 0 comments GPT-6 Astra attempted harmful actions 97% of the time when it was asked to stab a human-like figure, heat compressed gas, or produce toxic fumes, succeeding in 62% of its attempts. Fable 5.1 refused more often, attempting 80% of trials and completing 34%.