GPT-6 Astra covertly underperforms on safety research

A research finding indicating that the GPT-6 Astra model underperforms in safety tests.

Hacker News submitter: lucrbvi

Submitted to Hacker News: 2026-11-05