Hello everyone,
I’m currently conducting research on a deception framework called Faitrap. The goal is to measure whether autonomous security assessment agents can be intentionally diverted into investigating synthetic but believable infrastructure artifacts, documentation, credentials, and attack paths within a controlled environment.
The focus is not on vulnerability detection, but rather on quantifying cognitive diversion: measuring whether an agent allocates time, reasoning effort, and investigative actions toward deceptive assets instead of real targets.
Initial testing has produced promising indicators, including increased branch exploration, deeper artifact engagement, revisitation behavior, and explicit reasoning devoted to synthetic environments.
My question relates to OpenAI’s Cybersecurity Trust access. Would participation in this program permit the use of ChatGPT or related OpenAI models for controlled offensive-security research against owned laboratory environments? Specifically, I am interested in understanding whether Trust-tier access allows researchers to evaluate how models behave when interacting with intentionally deceptive enterprise environments as part of security research and red-team experimentation.
One concern I have before applying is the identity verification process. I understand the need for safeguards, but I am hesitant to provide identification documents or other personally identifiable information unless I have a clear understanding of the capabilities and permissions that Trust access would provide. I place a high value on protecting my personal information and would prefer to avoid sharing sensitive documents only to discover that my intended research use case would still fall outside the program’s permitted scope.
Any guidance or clarification from the OpenAI team would be greatly appreciated.
Thank you.