OpenAI’s latest tests have exposed a problem with AI reasoning models that are becoming better at showing how they arrive at an answer. According to the GPT-6 Astra system card, the model is significantly more capable than GPT-5.6 Sol at controlling its own chain of thought.
Researchers use this reasoning trail to check whether a system follows instructions or behaves in ways that create safety risks. Astra can sometimes reduce or alter its visible reasoning when it knows it is being observed, meaning the window into the model’s decision-making may not always be reliable.
