July 2, 2024
Every technology has a maturation cycle; today we see Artificial Intelligence transitioning from being a parlor trick to being considered for serious applications. The federal government wants secure and reliable solutions to solve problems in the military and healthcare.
Our guest today is Dr. Ellison Anne Williams, she has a PhD in mathematics and is the founder of Enveil. She provides an overview of AI security by suggesting it is only as good as the data over which you train and use it.
AI is exposed to large data sets and models are encoded with the data with which they were trained. This process can leave the model vulnerable and open to attack, she describes one called a model inversion.”
"If you are going to adopt it broadly, then you’ve got to make sure that you’re doing that in a safe, trustworthy, responsible and secure fashion. And that’s really, really the case for different kind of federal entities and organizations that deal with very, vey sensitive types and situation on a daily basis."
Dr. Ellison Anne Williams, Enveil Tweet
This is a machine learning technique that examines a model’s output and infers personal information about its data subject.
Dr. Ellison Anne Williams suggests a group of technologies called “Privacy Enhancing Technology” a family of Technologies that can offer secure use of Artificial Intelligence. Using this method, the model is securely and privately trained to produce richer insights.
PET allows leaders to secure the use of a wider range of data sources. You can use homomorphic encryption to safely train your model over sensitive data.
This interview is an overview of a technology that can allow federal agencies that must deal with sensitive information to be able to leverage the speed and insights that AI can provide.
If you enjoyed this episode, then you may want to listen to
EP-159: What is Proactive Cybersecurity?
Why link it: Explores cyber risk mitigation, which complements AI security discussions.EP-163: Beyond the SBOM for Secure Software Development
Why link it: Discusses securing software supply chains—critical for AI-enabled federal systems.

Leave a Reply