WALLIX and Inria Push Trusted AI to Strengthen Identity Security Amid Rising Machine Identity Risks
As machine identities proliferate across enterprise networks, the security assumptions that underpin traditional identity governance frameworks are being stress-tested by the scale and complexity of AI deployment. A collaboration between WALLIX and Inria, France’s national research institute for computer science, is tackling a critical question: How can identity security frameworks remain trustworthy when the systems they govern are becoming autonomous?
The Trust Equation in Agentic Environments
Identity governance depends on an implicit assumption: the identities we’re governing have known, predictable behavior. A user follows organizational policies. A service account executes its programmed task. But AI agents operate under a different behavioral model. Their actions emerge from neural network inference, making their behavior probabilistic rather than deterministic.
This creates a trust problem. How do you construct governance policies for identities whose next action is not predetermined? How do you audit access decisions made by entities that can explain their reasoning but not predict it in advance?
Trusted AI as an Identity Control Layer
WALLIX and Inria are approaching this by building what might be called “trusted AI governance”—using AI systems to monitor and validate the behavior of other AI systems. The concept inverts the traditional security model: instead of trying to restrict what AI agents can do through static policies, the research explores whether we can build AI systems that learn what “normal” agent behavior looks like and flag deviations.
This is distinct from traditional anomaly detection. The focus is not on detecting individual malicious actions but on understanding whether an AI agent’s behavior remains consistent with its trained objectives and role-based scope. An AI agent that suddenly begins querying systems outside its normal pattern, requesting excessive data, or operating at unusual hours would trigger governance alerts.
Machine Identity Governance at Scale
The collaborative research also addresses a practical NHI challenge: visibility. Enterprise networks often contain thousands of service accounts and machine identities. With AI agents being deployed at scale, the total count of non-human identities now exceeds human identities by orders of magnitude in many organizations.
Traditional identity governance tools were not designed for this scale. WALLIX’s identity and access management (IAM) platform combined with Inria’s machine learning expertise is exploring how to apply trusted AI techniques to the machine identity governance problem.
Building Governance Frameworks for Autonomous Enterprise Systems
The partnership represents an important recognition in the identity security industry: NHI governance at scale requires new models. As enterprises deploy AI agents for customer service, data analysis, and operational automation, the identity governance frameworks that protect those systems must evolve from static, provisioning-time control to dynamic, runtime monitoring.
Trusted AI governance offers a path forward—one where identity security remains trustworthy even as the entities it governs become increasingly autonomous.