OpenAI says Astra is its first model to reach its "Critical" cyber threshold and warns safeguards may mistakenly flag legitimate activity as cyber misuse (Ina Fried/Axios)

OpenAI has announced that its upcoming model, Astra, is its first to reach the "Critical" cyber threshold, indicating it can independently find and exploit previously unknown vulnerabilities in real-world software. According to OpenAI, this threshold is reached when a model poses new levels of risk, and the company has followed its procedure for this situation by halting further development until safeguards and security measures could be implemented.

Astra is said to be substantially more capable than OpenAI's current frontier AI model, GPT-5.6 Sol, particularly in cyber tasks. The company plans to release Astra soon, but only a handful of partners will have access to its most advanced cybersecurity capabilities. These partners, referred to as "alpha testers," include individuals and organizations responsible for protecting critical digital infrastructure, such as the U.S. government and companies in OpenAI's trusted access program for cybersecurity.

OpenAI has expressed concerns that its safeguards may mistakenly flag legitimate activity as cyber misuse. The company had previously paused some training workloads related to Astra's development, but has now resumed work after implementing additional safety and security controls. OpenAI sees its models being used to prevent cyberattacks, or for "defensive cybersecurity," as a critical revenue stream.

Brief written by urgent.news from Techmeme, Fortune, Wired — 3 reports on this story. Machine-written — may contain errors; check the original before relying on it.