Our Story
We have found exploits in devices used by millions of consumers and that are probably sitting in your homes right now. Building these skills takes years of practice and training, leading to a large disparity between the amount of firmware created today and the amount of people capable of securing those devices.
With the raise of AI we hoped that new models could help lessen this gap however the state of the art models today still can't handle this task. This is for a varity of reasons like models being general purpose and not specialized, models focusing on human readable language and not binary files, and more foundational AI problems such as sparse rewards and transfer learning.
Binary analysis provides a rich training ground for solving these problems and pushing the boundry of AI forward. You can easily generate large amounts of data and unique cases, experiment on different sizes of data, and work in a problem domain grounded in reality not a limited game world. We are pushing forward the boundries of AI in order to develop an agent that can exceed ourselves in detecting vulnerabilities in firmware.