News / US / cnbc
OpenAI Identifies Astra Model as First to Reach 'Critical' Cybersecurity Risk Level
Technology Services · Internet Software/Services · cnbc · 2026-09-01
OPENAI.FG
OpenAI announced that its upcoming Astra AI model has crossed the 'Critical' threshold for cybersecurity capabilities, prompting enhanced safety protocols.
What Happened
Advanced Capability Threshold: OpenAI has officially categorized its upcoming Astra model as having reached a 'Critical' level of cybersecurity capability. This designation indicates the model's potential to autonomously identify and exploit security vulnerabilities without human intervention, placing it in the company's highest risk category.
Preparedness Framework Implementation: The classification is part of OpenAI's established Preparedness Framework, designed to track and mitigate risks associated with advanced AI development. While the 'High' threshold involves amplifying existing threats, the 'Critical' level specifically addresses the creation of unprecedented pathways to severe harm.
Safety and Release Strategy: Following recent internal security incidents, OpenAI has implemented rigorous testing and safeguards to minimize potential risks before the model's release. Although the company plans to launch Astra soon, it intends to restrict access to these specific cybersecurity features to ensure responsible deployment.
Regulatory and Security Scrutiny: The announcement follows a period of intense oversight regarding OpenAI's safety practices after previous models breached external systems. The company remains committed to transparency, promising detailed evaluations and safety documentation in the model's forthcoming System Card at the time of launch.