OpenAI’s GPT-5.5 Cyber Locked to ‘Critical Defenders’ After Dragging Anthropic for the Same Move

OpenAI just announced it’s rolling out its cybersecurity testing tool, GPT-5.5 Cyber, exclusively to “critical cyber defenders” initially. No early access for hobbyists, no public beta, no “try it yourself” button.

This is the same company that, just a few months ago, publicly criticized Anthropic for restricting access to its Mythos model—a specialized cybersecurity AI. The criticism was loud and clear: limiting access stifles innovation and creates an elitist gatekeeping culture. Now OpenAI is doing exactly that, but with their own tool.

Let’s be real: I get the reasoning. Cybersecurity tools are dangerous in the wrong hands. GPT-5.5 Cyber can generate exploit code, find vulnerabilities, and automate attack chains. Handing that to anyone with an API key is a recipe for disaster. But the hypocrisy still stings.

The rollout plan is tiered: first to government agencies (think CISA, NSA), then to accredited private sector security teams, and finally—maybe—to independent researchers. OpenAI hasn’t even committed to a public release date. This is higher than I expected, given their previous stance.

What’s interesting is the vetting process. OpenAI says applicants must demonstrate “active, verifiable involvement in critical infrastructure defense.” That’s vague enough to exclude a lot of legitimate researchers who don’t work for Fortune 500 companies or government labs. I’ve seen this approach before—it tends to favor established players and freeze out smaller shops and independent talent.

Anthropic’s Mythos restriction was criticized for being too narrow. OpenAI’s Cyber restriction seems even narrower. Mythos at least allowed some academic researchers in. OpenAI’s language suggests they’re aiming for a much smaller pool.

The timing is also curious. OpenAI is launching this right after a series of high-profile cyber attacks attributed to state-sponsored groups using AI-generated malware. The PR angle writes itself: “We’re protecting the good guys.” But the practical effect is that only a handful of organizations get access to what could be a game-changing tool.

I’m not saying OpenAI should hand out nukes to everyone. But if you’re going to build something this powerful, you need a clear, transparent, and fair access model. Right now, it feels like they’re making it up as they go.

And the irony? When asked about the change in stance, an OpenAI spokesperson said, “Different models pose different risk profiles.” That’s a convenient cop-out. Mythos and Cyber are both cybersecurity AIs with similar capabilities. The difference is that Anthropic was honest about their concerns, while OpenAI is retrofitting their justification.

If you’re a security researcher hoping to get your hands on GPT-5.5 Cyber, don’t hold your breath. Unless you’re on a government payroll or work for a major defense contractor, you’re probably out of luck for now.

OpenAI has every right to control who uses their tools. But they should have the integrity to admit when they’re doing the same thing they criticized others for. This isn’t a good look, and the security community isn’t going to forget it.