OpenAI staff point readers to GPT-6 Astra system card's alignment evals

kaicathyc · x · 2026-09-04

An OpenAI deployment safety team member recommends reading the GPT-6 Astra system card beyond the launch blog: it shares extensive evals and analysis on the model's alignment, and discusses continued challenges on both alignment and monitoring fronts.

Related event: GPT-6 Astra System Card Flags Major Drop in Monitorability(37 posts)→

Original post →

More from Models

Models channel →