(POS-15) Utilization of Simple Badge Data for Insider Threat Detection via Artificial Neural Networks and Its Applications

Aidan Cook- Oak Ridge National Laboratory; Joe Beck – Oak Ridge National Laboratory; Debraj De – Oak Ridge National Laboratory; William Lorenzen – Harvard Children’s Hospital; John Landers – Oak Ridge National Laboratory; Scott Nelson – Oak Ridge National Laboratory

Insider threat continues to be a dynamic and emerging challenge for the security of nuclear and radioactive materials. As a result, control of radioactive sources and associated personnel presents several opportunities for adversarial interception or tampering with the sensitive materiel. To this end, leveraging data provided by onsite detectors, such as badge readers, can enable security and onsite personnel to enhance monitoring around these sensitive devices. By digesting these data via algorithms using artificial neural networks (ANNs), security personnel are given additional insight that can be used to detect potential insider threats. ANNs are a common tool in algorithms leveraging artificial intelligence (AI) and provide wide ranging insights to a variety of applications in transportation, packaging, supply chain, and beyond. Built upon the fundamental idea of our own biological neurons, an ANN takes a wide breadth of data from various sources to make informed decisions based on a large features space and inputs. Within this paper, we will discuss the risks presented by hosting category 1 and 2 sources related to insider threats and associated threat vectors. We will then explore how data that is commonly collected at various institutions, such as badge data, can be fed into ANNs for complex pattern-of-life and anomaly detection algorithms. Finally, we will disseminate the results of the proposed algorithm using the available data – this will expand into scalability internal and external to the facility. Specifically, we will discuss the enhanced security and insider threat identification and how it applies to other locations with similar data inputs.