OpenAI launched Trusted Access in February of this year as an identity and trust framework around its most cyber-capable frontier models. The problem it solves is that advanced cyber capabilities are dual-use by definition. “Hey Codex, find vulnerabilities in my code" is responsible security research from one keyboard and attack preparation from another, and it’s impossible to tell them apart from the prompt alone.
So OpenAI stopped gating requests and instead started vetting the requester. Octane is now on the list of approved organizations.
Earlier this month it was Anthropic's Cyber Verification Program. Now it's OpenAI's Trusted Access for Cyber. Both of the labs building the most capable models on Earth have examined what Octane does and concluded we belong on the short list of organizations trusted to do security research with their frontier models. Our customers don't have to bet on which lab wins, they get the strongest models from both labs running against their code.
We're now performing benchmarking and evaluation testing over these models: where they perform best, where they fall short, and how they can accelerate defensive security work. We'll publish what we learn. We're grateful to the OpenAI team for the access and the partnership.
Watch this space.
Subscribe to our newsletter
