OpenAI said it is introducing a new access framework for its latest cyber-focused artificial intelligence model, GPT-5.3-Codex, as part of an effort to widen the use of advanced AI in cybersecurity while limiting the risk of malicious activity.
The company said the model reflects a shift in AI capabilities, from assisting developers with small coding tasks to operating autonomously for extended periods to carry out complex security-related work, such as identifying and helping remediate software vulnerabilities.
See also: OpenAI Launches Frontier to Scale AI Agents in the Workplace
To manage the risks associated with these capabilities, OpenAI said it is piloting a trust-based system known as Trusted Access for Cyber, which relies on identity verification and organizational controls to prioritize access for defenders, researchers and enterprises focused on protection and resilience.
The framework includes technical safeguards such as safety training and automated monitoring designed to flag potentially harmful activity, while aiming to reduce friction for legitimate cybersecurity work that can be difficult to distinguish from misuse at an early stage.
See also: Google Rolls Out Gemini AI Features in Chrome to Boost Web Multitasking
Alongside the access pilot, OpenAI said it is allocating $10 million in API credits through its Cybersecurity Grant Program, targeting teams with experience securing open-source software and critical infrastructure, as the company seeks to encourage faster adoption of frontier AI tools for defensive cybersecurity efforts.
Source: OpenAI