OpenAI has announced a significant expansion of its Trusted Access for Cyber (TAC) program, alongside the launch of GPT 5.4 Cyber, a model specifically designed for defensive cybersecurity applications. This move is part of OpenAI's broader strategy to enhance cyber defenses while managing the risks associated with increasingly advanced AI models. The expansion aims to onboard thousands of verified individual defenders and hundreds of security teams tasked with protecting critical software and infrastructure.
The Trusted Access for Cyber program, initially introduced earlier this year, is designed to provide vetted cybersecurity professionals with controlled access to advanced AI tools. These tools, due to their dual-use nature, are typically restricted. OpenAI is now introducing additional access tiers based on identity verification and trust signals, allowing individual users to verify themselves through structured onboarding and enterprises to request access for their teams. This approach shifts away from manual access decisions to a more objective verification method.
A key element of this expansion is the introduction of GPT 5.4 Cyber, a specialized version of OpenAI's latest model fine-tuned for cybersecurity tasks. Unlike general-purpose models, GPT 5.4 Cyber is more permissive with cyber-related queries, enabling security professionals to conduct tasks such as binary reverse engineering and vulnerability analysis. However, access to this model is limited and controlled, reflecting concerns about its dual-use potential.
OpenAI's cybersecurity strategy focuses on democratized access, iterative deployment, and ecosystem resilience. The company acknowledges the dual-use nature of AI tools, which are increasingly used by both defenders and attackers. To address this, OpenAI combines broader access to general models with stricter controls for advanced capabilities, requiring stronger verification and accountability from users.
The expansion of the Trusted Access for Cyber program highlights the need for a structured model for deploying sensitive capabilities. By focusing on verification and trust-based access, OpenAI aims to provide defenders with powerful tools while mitigating the risk of misuse. The controlled rollout of GPT 5.4 Cyber underscores the ongoing challenge of balancing access with security in the evolving field of cybersecurity.
Source: https://openai.com/index/scaling-trusted-access-for-cyber-defense/