OpenAI just announced it's rolling out GPT-5.5 Cyber with restricted access, limiting the tool to "critical cyber defenders" initially. If that sounds familiar, it's because OpenAI recently called out Anthropic for doing essentially the same thing with their Mythos model.
The irony here is hard to miss. OpenAI positioned itself as more open and accessible when Anthropic limited Mythos access, but now they're taking the exact same approach with their cybersecurity tool.
To be fair, there's a legitimate argument for restricting access to powerful cybersecurity AI. These tools can identify vulnerabilities and potentially be misused by bad actors. Starting with trusted security professionals makes sense from a safety perspective.
For AI practitioners, this signals a broader trend. As AI capabilities grow more powerful in sensitive domains like cybersecurity, expect more gated rollouts regardless of which company is doing the releasing. The days of immediate public access to cutting-edge models may be ending.
The real question is whether OpenAI will acknowledge the contradiction or quietly hope everyone forgets their earlier criticism of Anthropic's approach.
