OpenAI has granted limited access to its new AI model, GPT-5.4-Cyber, to a select group of users. This model is marketed as an effective solution for identifying vulnerabilities in software.
We’re expanding Trusted Access for Cyber with additional tiers for authenticated cybersecurity defenders.
— OpenAI (@OpenAI) April 14, 2026
Customers in the highest tiers can request access to GPT-5.4-Cyber, a version of GPT-5.4 fine-tuned for cybersecurity use cases, enabling more advanced defensive workflows.…
OpenAI is expanding its Trusted Access for Cyber program launched in February. This initiative involves verifying individuals and collaborating with a limited number of organizations to provide models with "more liberal restrictions in the field of cybersecurity."
Additional access tiers have been introduced for users wishing to validate their qualifications as cybersecurity specialists.
Clients with the highest status will gain access to GPT-5.4-Cyber—a specialized model with enhanced capabilities and fewer basic restrictions.
"This is a version of GPT-5.4 that lowers the barrier for legitimate work in cybersecurity and unlocks new capabilities for advanced defensive workflows," the announcement states.
Among the new features of the neural network is the ability to reverse engineer binary files. This allows experts to analyze compiled software for vulnerabilities and malware, as well as assess overall security levels without direct access to the source code.
The tool was launched a week after the high-profile announcement of Mythos. Anthropic decided against releasing the model to the public due to its advanced capabilities in vulnerability detection and high security risks.
To use GPT-5.4-Cyber, users must verify their identity, while organizations need to request access through their representative.
In April, OpenAI announced a plan aimed at strengthening child safety in an era of widespread AI use and technology misuse.