OpenAI will restrict access to GPT-5.5 Cyber, its new offensive-security model, to vetted 'critical cyber defenders,' Sam Altman said in a post on X on Thursday, April 30, 2026. The rollout begins in the next few days. The decision mirrors the exact gating strategy Altman publicly mocked Anthropic for adopting with its rival tool, Mythos.
Would-be users have to fill out an application on OpenAI's website, describing their credentials and what they plan to do with the model. The form implies Cyber can perform penetration testing, vulnerability identification and exploitation, and malware reverse engineering. In other words, it is a packaged offensive-security toolkit, sold as a way for companies to find their own holes before someone else does.
When Anthropic limited Mythos to a small set of approved users earlier this cycle, Altman called the move 'fear-based marketing.' Some outside critics agreed, arguing Anthropic's framing of Mythos as too dangerous for general release was overblown and conveniently good for the brand. The reversal at OpenAI lands without much explanation of what changed.
Key facts
- 01OpenAI will roll out GPT-5.5 Cyber only to vetted 'critical cyber defenders' starting in the next few days, Sam Altman said April 30, 2026.
- 02Access requires an application on OpenAI's site detailing credentials and intended use.
- 03Cyber is pitched for penetration testing, vulnerability identification and exploitation, and malware reverse engineering.
- 04Altman previously called Anthropic's identical gating of its Mythos tool 'fear-based marketing.'
- 05An unauthorized group reportedly obtained access to Mythos anyway after Anthropic restricted it.
There is also the awkward fact that Anthropic's gating did not actually contain Mythos. An unauthorized group reportedly obtained access to the tool anyway, undermining the premise that a credential check at the front door keeps a capable cyber model out of the wrong hands.
“Altman called Anthropic's Mythos gating 'fear-based marketing,' then confirmed OpenAI will roll GPT-5.5 Cyber out only to 'critical cyber defenders' in the next few days.”— Jaeden Schafer
OpenAI says it is working to widen access over time by consulting with the U.S. government and by identifying more users with legitimate cybersecurity credentials. That suggests a tiered rollout closer to how export-controlled software is distributed than how ChatGPT itself reached hundreds of millions of users.
The pivot is notable because OpenAI has historically leaned toward broader, faster availability than Anthropic, which has built much of its public identity around safety-first deployment. On Cyber, the two labs have converged on the same answer: a model that can write working exploits is not something you ship behind a credit card form.
It also lands in a busy week for OpenAI's security posture. The company announced new advanced security features for ChatGPT accounts, including a partnership with Yubico for hardware-key authentication. Separately, Elon Musk testified that xAI trained Grok on OpenAI models, a reminder that even well-guarded model weights leak influence into competitors' systems.
The case against gating is the one Altman himself made weeks ago. Defenders need these tools as much as attackers do, and a slow application queue at OpenAI does not help a mid-sized company whose cPanel installation is being actively exploited today. Restricting Cyber to a credentialed elite risks widening the gap between well-resourced security teams and everyone else, while doing little to stop a determined adversary who, as the Mythos leak showed, can get the model anyway.
The case for gating is harder to dismiss now that the capabilities are real. A model that can chain reconnaissance, exploit development, and malware analysis end-to-end is materially different from a chatbot that writes Python. Once that capability is generally available through an API, the marginal cost of running a campaign collapses.
Altman's earlier broadside at Anthropic looks, in hindsight, like a marketing position he could not hold once OpenAI's own offensive-security model was ready to ship. The substantive question, which neither company has answered cleanly, is whether credential-gated access plus government consultation is actually a security control or just a liability shield. The Mythos breach suggests the former is harder than the press releases imply, and OpenAI is about to run the same experiment in public.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
