Trusted access for the next era of cyber defense

We are scaling up our Trusted Access for Cyber (TAC) program to thousands of verified individual defenders and hundreds of teams responsible for defending critical software. For years, we’ve been building a cyber defense program on the principles of democratized access, iterative deployment, and ecosystem resilience. In preparation for increasingly more capable models from OpenAI over the next few months, we are fine-tuning our models specifically to enable defensive cybersecurity use cases, starting today with a variant of GPT‑5.4 trained to be cyber-permissive: GPT‑5.4‑Cyber. In this post, we share how we expect our approach of scaling cyber defense in lockstep with increasing model capabilities to guide the testing and deployment of future releases.

The progressive use of AI accelerates defenders – those responsible for keeping systems, data, and users safe – enabling them to find and fix problems faster in the digital infrastructure everyone relies on. Similarly, AI is being used ⁠ by attackers looking to cause harm. We’ve been preparing for this. Since 2023, we’ve supported defenders through our Cybersecurity Grant Program ⁠ and strengthened safeguards through our Preparedness Framework ⁠ . The same year, we started evaluating our models' cyber capabilities, and in 2025, we began including cyber-specific safeguards ⁠ (opens in a new window) in our model deployments ⁠ . Earlier this year, we furthered our support for defenders with the launch of Codex Security ⁠ to identify and fix vulnerabilities at scale. Our approach to this continuous advancement of capabilities is guided by three principles:

Democratized access: Our goal is to make these tools as widely available as possible while preventing misuse. We design mechanisms which avoid arbitrarily deciding who gets access for legitimate use and who doesn’t. That means using clear, objective criteria and methods – such as strong KYC and identity verification – to guide who can access ⁠ more advanced capabilities and automating these processes over time. Ultimately, we aim to make advanced defensive capabilities available to legitimate actors large and small, including those responsible for protecting critical infrastructure, public services, and the digital systems people depend on every day.

Cyber risk is already here and accelerating, but we can act. Digital infrastructure has already been vulnerable ⁠ (opens in a new window) for years, before advanced AI even came along. Now, existing models can help find vulnerabilities, reason across codebases, and support meaningful parts of the cyber workflow, and threat actors are experimenting with novel AI-driven approaches. We’ve seen sophisticated harnesses elicit stronger and stronger capabilities by using more test-time compute with existing models. That means safeguards cannot wait for a single future threshold.

To enable responsible use at scale, we need systems that can validate trustworthy users and use cases in more automated and more objective ways. This allows us to expand access based on evidence and real signals of trust, rather than relying on manual decisions. We don’t think it’s practical or appropriate to centrally decide who gets to defend themselves. Instead, we aim to enable as many legitimate defenders as possible, with access grounded in verification, trust signals, and accountability.

Investing in ecosystem resilience: We support and accelerate the community of defenders through trusted access pathways, targeted grants ⁠ , contributions to open-source security initiatives ⁠ (opens in a new window) , and technologies like Codex Security ⁠ that help defenders more rapidly find and patch vulnerabilities.

We began cyber-specific safety training with GPT‑5.2, then expanded it with additional safeguards through GPT‑5.3‑Codex and GPT‑5.4, where we also classified the model as “high” cyber capability under our Preparedness Framework. In parallel, we increased support for defenders: launching a $10M Cybersecurity Grant Program ⁠ , reached over 1,000 open source projects with Codex for Open Source ⁠ (opens in a new window) which provides free security scanning, and continued to improve Codex Security.

All customers approved through this process will gain access to versions of existing models with reduced friction around safeguards which might trigger on dual-use cyber activity, allowing them to continue to support security education, defensive programming, and responsible vulnerability research. Customers already in TAC willing to further authenticate themselves as legitimate cyber defenders can express interest ⁠ (opens in a new window) in additional tiers of access, including requesting access to GPT‑5.4‑Cyber.

Iterative deployment: We learn the most by putting these systems into the world carefully ⁠ and improving them over time. As we better understand both their capabilities and risks, we update our models and safety systems accordingly. This includes understanding the differentiated benefits and risks of specific models, improving resilience to jailbreaks and other adversarial attacks, and improving defensive capabilities — while mitigating harms.

For years, our cybersecurity strategy has been to invest in research, prevent misuse, and accelerate defenders. As model capabilities have advanced, we have expanded our programs toward these goals, which are grounded in the following convictions:

Expand access based on who is using these systems and how they’re being used. Cyber capabilities are inherently dual-use, so risk isn’t defined by the model alone. It also depends on the user, the trust signals ⁠ (opens in a new window) around them, and the level of access they’re given.