On September 4, 2026, OpenAI announced the launch of its latest artificial intelligence model, GPT-6 Astra, claiming it to be “the most intelligent and aligned model in the world,” with significant improvements in cybersecurity capabilities. However, the model has raised widespread concerns in the security domain due to the lack of direct visibility into its internal reasoning process. These concerns intensified after Hugging Face’s recent hacking incident, which led to a Chinese open-source model being involed in the investigation. Greg Brockman, President of OpenAI, made a statement on this matter during a press conference on Thursday.
The GPT-6 Astra model launched by OpenAI lacks direct visibility into “how to think,” and this development has raised security concerns. These concerns surfaced several weeks after the Hugging Face hacking incident, which forced a Chinese open-source model to step in for investigation. When announcing Astra on Thursday, OpenAI described it as “the smartest and most aligned model in the world” and claimed that it “has made significant progress in cybersecurity capabilities.” Greg Brockman, President of OpenAI, said at the press conference on Thursday...