An end-to-end adversarial input generator

UT-Battelle, LLC, acting under its Prime Contract No. DE-AC05-00OR22725 with the U.S. Department of Energy for the management and operation of the Oak Ridge National Laboratory (ORNL), is seeking a commercialization partner for new end-to-end adversarial input generator . The ORNL Technology Transfer Office will accept license applications through June 21, 2022. The Problem: Adversarial samples are a new class of malware specifically engineered to evade classification by machine learning—based malware detection systems. These samples constitute a serious threat because they can transfer between models, i.e., adversarial samples developed for use against a particular model architecture can achieve high success rates against completely different architectures, without needing to invest significant time and resources. Despite the threat this class of malware poses, no end-to-end systems for detecting and resolving vulnerabilities across the entire spectrum of semantic alterations exist. The Solution: Researchers at ORNL have developed an end-to-end adversarial input generator that consumes a database of malware files and outputs a database of adversarial samples. The developed technologies have been implemented as a modular system that not only incorporates major types of semantic alterations but also allows new alteration types to be added in as they are developed. Applications: Evaluation and enhancement of new or existing cyber security tools and products; Development of robust machine learning Advantages: Capable of generating large quantities of evasive malware samples; Can detect malware detection system weaknesses, as opposed to generic vulnerabilities that are common across the industry; Produces a report of evasive samples that bypassed the tested tool and what alteration worked; Immediately provides tested evasive samples to vulnerability detection for retraining detection models License applications will be evaluated based on prospective partners' ability and commitment to successfully commercialize the technology, with a preference for United States-based businesses and small businesses. For a license application, contact Andreana Leskovjan, Commercialization Manager, [email protected]. ORNL Intellectual Property: ID 4695: Functionally-Preserving Adaptive Malware Binary Rewriting Copyright ID 90000221: AMIGO