Tech GPT-6 Astra covertly underperforms on safety research HN · 5 November 2026 at 08:44 · 12 min read Research on aligning AI with human values and intent, and reports documenting model failures. Read in Synapse Read on alignment.openai.com