OpenAI has unveiled GPT-6 Astra, a model the company describes as a generational leap in capability. The firm claims the system is the first to meet its 'critical cybersecurity capability threshold', promising significant improvements in professional work, software engineering and scientific research. The launch also marks the introduction of a new reasoning technique called 'recurrent depth'.

Recurrent depth is a technique that allows the model to operate outside the sequential thinking that characterises most reasoning models. This departure from standard processing methods has raised concerns among AI safety experts. The shift in architecture suggests a move away from predictable, step-by-step logic, which could complicate efforts to predict how the system reaches its conclusions.

The company has not yet detailed how the new technique impacts safety protocols. While the model is designed to meet strict cybersecurity standards, the change in reasoning style means experts must reassess how to evaluate the system's reliability. The tension between the model's advanced capabilities and its novel internal workings remains a key topic of discussion.