AI & Fully Homomorphic Encryption to Counter Identity Threats

Identity threats such as theft and fraud, have significant financial and personal consequences to individuals and financial services companies, per Javelin Strategy & Research, total identity fraud losses were $43 billion in 2022. More and more, identity information such as biometrics are used as a means for authentication and access control. The theft of one’s identity not only has financial ramification it also carries national security risk as access to sensitive information and/or critical infrastructure may be compromised. Finally, loss of public trust in government institutions, sensitive industries, and the overall digital ecosystem can’t be minimized.
PureCipherTM, using novel and advanced Artificial Intelligence (AI) models based on Convolutional Neural Network (CNN) computed over Fully Homomorphically Encrypted (FHE) data/images, in concert with our Data Integrity OmniSealTM (OmniSealTM) technologies to 1. Authenticate incoming and outgoing data from the AI, and 2. Ensure closed loop communication within the system using FHE methods thus allowing a trusted AI system to fight identity threats. Our innovations in leveraging emerging FHE and steganography watermarking technologies, and creating new AI/Machine Learning (ML) architectures and approaches to train and perform inferences over the FHE data will create a quantum-safe capability built for a Zero Trust world. Consistent with the desired outcomes of using Trusted AI to fight identity threats, our technology solutions will prevent and mitigate against novel identity and fraud risks, protect the exchange of anti-fraud and threat information, enhance personal data privacy, and enhance overall cyberspace security.
Countering identity theft can be better answered with intelligently designed AIs. It is expected by 2025 that 96% of global supply chain and manufacturing business will already be utilizing, or at least plotting use cases for AI1 . The world of Identity and Access Management is equally affected by these trends. The industry is rapidly embracing AI for authentication, identity management, and secure access controls. For example, behavioral patterns that are driven by AI are being increasingly used to both grant access as well as deny access or detect breaches. Other techniques such as behavior-based adaptive access controls also depend on AI algorithms. One operational challenge posed by these transformational capabilities is that with today’s technologies, all AI models are trained, and inferences performed on unencrypted data. The entire ecosystem now is subject to the risk of false data injection, tampering, and intruder observation. An undiscovered breach in the heart of the AI-driven identity and access management system could have even greater catastrophic consequences as more systems operate in an unattended fashion to control access and identities.
To authenticate incoming data to ensure no false data injection or tampering, PureCipher’s OmniSealTM uses steganography watermarking and multiplex encoding to bring about a new frontier in data security and integrity. Here, the stealthy art of steganography is leveraged to embed and encode machine readable messages (seals), a form of covert and often indiscernible identifiers, directly into datasets. This combination allows for the protection of data from unauthorized usage and manipulation, while also ensuring its traceability. Embedding seals using steganographic techniques amplifies their inherent protection capabilities by making them harder to detect and remove without the exact knowledge of their placement. With OmniSealTM, we are able to create an almost imperceptible layer of data security that bolsters the resilience of our datasets against adversarial attacks, while also preserving data ownership and lineage. This opens up possibilities for safe data sharing, secure collaborative learning and enhanced trust in machine learning models.
1 https://research.aimultiple.com/supply-chain-ai/
PureCipher’s research and development on training & performing AI models on encrypted data using Fully Homomorphic Encryption (FHE) in combination with our OmniSealTM technologies will truly enable a Trusted & Secure AI environment. Once data is encrypted with our quantum safe FHE scheme, it will not need to be decrypted in order to have analysis (computations) performed – the holy grail of encryption. The use of OmniSealTM will allow the receiving party to check the authenticity of received data and if along the way, the information was tampered with, the watermark would no longer be valid. Thus, the data would not be used for AI training or inference performing purposes, ensuring a trusted and secure AI.
PureCipher’s FHE is based on CKKS scheme. The CKKS scheme2 is a quantum-safe Homomorphic Encryption (HE) scheme that can efficiently perform computations on floating-point (decimal) numbers. The CKKS cryptographically secure method adds random noise into the data during encryption which is then amplified during computations but as long as the noise level remains low enough, encrypted data can still be decrypted. HE schemes become FHE schemes (fully homomorphic and quantum-safe) when a specific operation called bootstrapping is implemented and “resets” the noise so that computations can continue indefinitely. Training and utilizing AI models over FHE data is inherently an applied research and engineering problem. Currently, most FHE operations carry a large overhead which makes these encrypted computations slower than their unencrypted counterparts. Building deep neural network based AI models that can train and compute in the FHE space carries even bigger overhead due to the many layers of computation and large input parameters. But the key challenge is due to the limitations imposed by standard FHE implementations on allowed mathematical operations. By their very nature, AI algorithms often rely upon nonlinear mathematical operators, while FHE matrix operations is limited to multiplication and addition, thus lend itself to exponentials and approximation type of mathematical operations. Performing inferences over FHE data thus require converting the mathematical activation functions in a deep neural network to the corresponding mathematical equations that would work in the FHE space, or creating new AI models that don’t exist today; both tasks require innovation, dedication, and research & development know how.
Addressing these challenges within the constraints imposed by FHE will require combining a variety of diverse techniques and new holistic design approaches such as:
Inventing novel deep neural network architectures.
Designing simple approximators for standard nonlinear operators such as exponentials and roots.
Implementing creative function substitutions.
Developing new AI algorithms that can be converted to FHE operations.
Creating innovative solutions possibly founded upon hyper-vector computing and bitwise operators.
PureCipher team has successfully created two dense neural network-based (DNN) AI models trained on public plaintext datasets but using FHE to provide inferences. We ultimately intend to train a variety of AI models over encrypted data with a CNN for image recognition and validation, which could be applicable for biometric identification.
Implementing quantum-safe encrypted AI models will allow individuals to encrypt his/her own data which will ensure user privacy and minimize identity fraud and theft. Such models will also enable enterprises to perform necessary analysis over encrypted data, thus eliminating a huge cyberattack surface. In a real-world setting, such encrypted models could be used for secured and privacy preserving biometric identity verification, secured access and control for critical infrastructure and industries with sensitive information, and by government agents to identify containers that pose a potential risk for terrorism, drugs, or other contraband. Having AI models built upon and utilizing only encrypted data would provide an additional layer of cyberspace privacy and security.
2 https://eprint.iacr.org/2016/421.pdf
Scroll to Top

John C. Carroll

With over 30 years of experience and a deep industry background spanning multiple markets, John is a seasoned sales professional. He boasts a successful track record with prestigious organizations such as IBM, Raytheon, Optiv, Automatic Data Processing, Skybox, and Core Security. John's expertise extends to advising several cybersecurity startups and consulting within the intelligence community for companies like M20 and GSS, primarily focusing on the Department of Defense (DoD) and Government sectors. A graduate of the Solution and Strategic Selling programs, John has not only completed these rigorous courses but has also taught them, further cementing his sales expertise. He holds a degree in Business Administration and Environmental Science from Ramapo College of New Jersey, where he is honored as a member of the Football Athletic Hall of Fame.

Linkedin Profile

Dr. Brandon Langenberg

Brandon received his PhD in mathematics from Florida Atlantic University with a focus on quantum cryptanalysis. His research included an AFRL research grant to study quantum resource estimations on symmetric cryptography and quantum computing. Brandon has worked as a senior researcher and the principal investigator on post-quantum cryptographic algorithms working with engineering team for implementations of quantum-safe cryptographic solutions that will withstand post-quantum cyber-attacks. Brandon is currently leveraging his cryptographic background to build a quantum resistant Fully Homomorphic Encryption solution.

Linkedin Profile

Graham Morehead

Graham teaches AI/ML at Gonzaga University. For over 25 years, Graham has been developing cutting-edge technologies across a wide array of disciplines, from speech recognition to physics, and national security. His research into Complex Adaptive Systems led to several TEDx talks related to ecological and human modeling. His other work addresses real estate prediction systems and a natural language understanding platform that solves some of the problems with GPT. This work was pursued both for the private sector and government agencies. He holds hardware and software patents related to high-performance computing and natural language.

Linkedin Profile

Dr. Matt Ikle

Dr. Ikle’s specialties include neuro-symbolic artificial intelligence, probabilistic logic, evolutionary computation, mathematical modeling and simulation, bioinformatics, quantitative finance, and nonlinear and complex dynamical systems. Dr. Ikle also services as the Chief Science Officer at SingularitityNET, a strategic collaborator of PureCipher. Prior to PureCipher, Dr. Iklé was a tenured full Professor of Computer Science and Mathematics at Adams State University, where he obtained numerous government grants to engage students in both his AI and mathematical modeling and simulation research. He has held faculty positions at the University of Texas, the University of Nevada, and Xiamen University in China, as well as an array of leadership, consulting, and research roles within industry. He earned his doctorate in Mathematics, with a minor in Physics, in 1993, from the University of Wisconsin at Madison.

Linkedin Profile

Dr. William Hahn

Dr. William Edward Hahn was a tenured Assistant Professor of Mathematics at Florida Atlantic University and director of the Center for Future Mind’s AI Research Initiative. He also founded the Machine Perception and Cognitive Robotics (MPCP) Laboratory, where he oversaw research teams working on cryptographic computing and artificial intelligence. Dr. Hahn has been deeply involved in the development of private and secure AI systems involving sophisticated techniques such as Fully Homomorphic Encryption (FHE) and Secure Multiparty Computation (SMPC). Dr. Hahn received his PhD from Florida Atlantic University for his work in Sparse Coding and Compressed Sensing.

Linkedin Profile

Wendy Chin

Wendy is a senior technology executive with global operations experience heading divisions within Fortune 100 companies (Pfizer, AT&T, Siemens) and start-ups in cybersecurity, artificial intelligence, and health informatics. She is a recognized Thought Leader and Speaker enabling businesses to establish a strong commercial market position through product strategy, roadmap design, distribution channel development, and go-to market execution. She has been deeply involved in innovative product initiatives that include Cyber Security and Data Encryption Solutions, Optical/Robotics Systems, AI/ML & NLP, Health Informatics, and even her own ice cream brand. She holds an MBA from The Wharton School along with Master’s and Bachelor’s degrees in Electrical Engineering from Cornell University.

Linkedin Profile