OpenAI has commenced a limited release of its new GPT-5.5-Cyber model, making it available exclusively to a select group of "cyber defenders." This rollout comes merely weeks after CEO Sam Altman publicly criticized Anthropic for adopting a similar restrictive approach.
Altman announced on X that the deployment would begin "in the next few days," with access reserved for trusted defenders focused on securing critical systems. He stated that OpenAI intends to collaborate with the broader ecosystem and government to establish "trusted access for cyber," aiming to "rapidly help secure companies and infrastructure."
GPT-5.5-Cyber is engineered to identify system vulnerabilities before they can be exploited. OpenAI asserts its capabilities in penetration testing, bug discovery, exploitation, and malware analysis. However, a common concern is that tools capable of system compromise rarely remain solely in benevolent hands.
The irony is notable, given Altman's previous remarks on the Core Memory podcast, where he critiqued Anthropic's analogous strategy of restricting access to its Claude Mythos model (provided to approximately 50 organizations and not publicly available). He framed Anthropic's approach as "exclusivity dressed up as caution," likening it to "selling fear" – "We have built a bomb, we are about to drop it on your head. We will sell you a bomb shelter for $100 million." OpenAI now appears to be implementing its own access controls.
Independent assessments validate the model's efficacy beyond marketing. The UK's AI Security Institute recently declared GPT-5.5-Cyber "one of the strongest models we have tested on our cyber tasks," noting it is only the second system to successfully complete one of its multi-step attack simulations from start to finish.
While positioned as a protective measure, when tools possess both offensive and defensive capabilities, the decisive factor often hinges on who gains access and utilizes them first.