Introducing Trusted Access for Cybersecurity

OpenAI has unveiled GPT-5.3-Codex, its frontier reasoning model with the strongest cyber capabilities to date. This model goes beyond simple code completion to autonomously carry out complex tasks, dramatically speeding up the discovery and resolution of vulnerabilities and strengthening the cyber defense ecosystem.

However, such powerful capabilities also carry the risk of being exploited. To help security professionals use the model smoothly while preventing malicious use, OpenAI is piloting a trust-based access framework called Trusted Access for Cyber. This approach aims to provide low-friction access for defenders through identity verification, while blocking prohibited actions such as data exfiltration or malware generation.

The key approaches are as follows.

• Identity verification: Users can verify their identity through a dedicated page.
• Enterprise access: Enterprises can request trust-based access for their entire teams through an OpenAI representative.
• Invite-only program: A separate program is operated for security researchers, granting access to a higher-privilege version of the model.

In addition, to expand the cyber defense ecosystem, OpenAI is providing $10 million in API credit through the Cybersecurity Grant Program. This program targets teams that contribute to identifying and resolving vulnerabilities in open source software and core infrastructure systems.