.png&w=3840&q=75)
AI-Driven Dynamic Access Control for IAM in Zero Trust Architecture
A Zero Trust IAM prototype that analyzes user, device, network, geolocation, and behaviour signals to classify access as Allow, Step-Up MFA, or Deny in real time.
The Problem
Static access policies cannot adapt to anomalous behaviour. A Zero Trust system needs to continuously evaluate signals and step up authentication when risk is detected.
The Solution
Built a Zero Trust IAM prototype that analyzes user, device, network, geolocation, keystroke, and session behaviour signals to classify access as Allow, Step-Up MFA, or Deny. Generated a 50,000-row synthetic IAM dataset with 26 attributes and 1,400 users. Used Isolation Forest and Variational Autoencoder models exported to ONNX for low-latency inference.
The Outcome
Achieved 0.91 F1-score, 97% simulated attack blocking, 92% false-positive reduction, and 48 ms average decision latency.
Project Details
A Zero Trust IAM prototype that analyzes user, device, network, geolocation, keystroke, and session behaviour signals to classify access as Allow, Step-Up MFA, or Deny. A 50,000-row synthetic IAM dataset with 26 attributes and 1,400 users was generated for training. Models include Isolation Forest and a Variational Autoencoder, exported to ONNX for low-latency inference via REST API and gRPC. Integrated with Azure AD Conditional Access. Achieved 0.91 F1-score, 97% simulated attack blocking, 92% false-positive reduction, and 48 ms average decision latency.
Technologies
Want something similar?
Post your project requirements and we'll build it for you.
Request a similar project