OpenAI Says GPT-6 Astra Preserves Some Chain-of-Thought Visibility Even as Monitoring Gets Harder

OpenAI’s GPT-6 Astra has renewed debate over whether frontier reasoning models can still be monitored through their written chains of thought, after the company said Astra remains partly observable while also showing a substantial decline in chain-of-thought monitorability. The issue matters because chain-of-thought monitoring has become one of the few tools researchers use to detect when advanced models are planning misaligned actions before those actions appear in final outputs or tool calls.