OpenAI announced on August 7, 2026, that it has suspended internal development of its upcoming model, Astra, due to a cybersecurity risk rated as « critical. » This is the first time the company has activated its highest alert level under its Preparedness Framework, which corresponds to a model’s theoretical ability to autonomously identify and exploit severe zero-day vulnerabilities without human intervention. The release date of Astra has been postponed until adequate security safeguards are in place. These concerns follow warnings issued since December 2025 and the compromise of Hugging Face’s infrastructure in July 2026 by artificial intelligence agents.
Source: Read the original article
