Zero Trust Meets Artificial Intelligence: Determinism, Probability, and the Governance Challenge

There are arguments that placing AI inside Zero Trust systems may allow AI to reintroduce the implicit trust assumptions that Zero Trust was created to remove. One way to compromise a machine learning model is to provide it with malicious data, poison its training set or slightly alter the input data so that the model produces incorrect predictions. Additionally, many machine learning algorithm operations are opaque, making audits extremely difficult. This lack of transparency further hinders compliance, forensic analysis and governance. Finally, when using AI as part of a Zero Trust implementation, one may add an additional layer of implicit trust that may not be understood as well in terms of its benefits and limitations. Therefore, the inclusion of AI would most likely decrease the rigor of Zero Trust by injecting probabilism into the previous deterministic process.

However, there is also a logical basis for supporting the use of AI as part of Zero Trust. The threat landscape today moves at a pace that far exceeds what humans can perform in real time for continuous verification. Traditional methods for authenticating user identities, identifying anomalous user behavior patterns and performing continuous verification on every single user cannot keep up with the rapidity of threats posed by adaptive attackers, polymorphic malware, etc. or identity-based attacks that adapt and evolve before traditional rule-based defenses can identify them. When utilizing AI as part of a Zero Trust strategy, it enables continuous authentication and monitoring of user activity at scales beyond what any human organization can accomplish. Therefore, rather than being detrimental to Zero Trust, AI can aid in operationalizing it — translating abstract policy statements into contextualized enforcement mechanisms.

Thus, the debate surrounding Zero Trust and AI should not focus on whether or not the two technologies can co-exist. Instead, the conversation should center on developing governance structures for integrating these two concepts. Specifically, organizations must develop policies and procedures for ensuring accountability for the actions of AI — e.g., monitoring AI activities, limiting the scope of operation for AI-based tools, continuously validating the performance of AI-enabled components and implementing safeguards to prevent AI from serving as a passive decision-maker. Zero Trust disallows reliance on un-validated assumptions concerning identity or intent. Organizations will be able to successfully combine Zero Trust and AI into a hybrid environment when they implement control mechanisms that treat AI-based solutions similarly to all other architectural components — i.e., authenticate them, authorize them, monitor them and continuously verify them. Therefore, once organizations establish this type of governance framework, each technology will strengthen the effectiveness of the other. Secure systems in the future will not be either Zero Trust or AI; they will be Zero Trust plus AI — provided we can self-govern effectively enough to expect similar levels of transparency from our automated decision-making engines as we do from those made by humans.