OpenAI Restricts Cybersecurity Features for Next-Gen AI Model Astra

OpenAI has restricted access to the cybersecurity features of its next-generation AI model Astra, implementing tighter security controls ahead of its release. The company stated that Astra has reached a 'critical' cybersecurity capability threshold under its Preparedness Framework, meaning it can identify and exploit unknown vulnerabilities in protected systems without specific human instructions if given proper tools and permissions.
OpenAI announced on January 1 that it has delayed parts of Astra's development and release over the past week to test protective measures against cyber misuse and unauthorized model behavior. Advanced cybersecurity functions will initially be available to early trial users, with broader defensive use cases planned for later rollout. The company also plans to publish detailed safety, security, and alignment assessments for Astra in its system card at launch.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: 오픈AI, GPT-6 아스트라 사이버보안 통제 강화