Hear from our internal security team as they share how we leverage User Entity Behavioral Analytics (UEBA) and a UCI-based confidence score to flag the moment an agent's behavior drifts from baseline.
Learn how to:
- Apply UEBA to build a behavioral fingerprint for an AI agent and detect when it's been compromised or gone rogue
- Recognize the risky agent behaviors UEBA flags (spikes in downloads/uploads, DLP policy violations, first-time access to new applications, and spikes in failed logins)
- Interpret the User Confidence Index (UCI) score to gauge an agent's risk level and take timely action
- Combine UEBA with traditional controls (access management, contextual DLP, and human-in-the-loop approvals) for defense in depth against agentic AI risk
For more information, check out our blog post.
View past events in this series!
Check out some customer questions below, or feel free to comment and continue the discussion!
Q: We are using Anthropic Claude AI within our environment. We would like to understand how it can assist with our day-to-day activities and provide visibility into user interactions and activities performed through the platform.
A: Different Netskope products like DLP, AI Security Guardrails, and AI Gateway would provide visibility.
Q: Can the UCI score be reset in case the detection is false positive?
A: Yes, this can be done with the help of an API endpoint.
Q: Is there a way to prevent users login to AI Agents using their personal mail and only login using a corporate one?
A: Yes, constraints for domains can be used in realtime policies to achieve this.
Q: How can we monitor and track the prompts entered by users and the attachments uploaded to any AI system within our environment?
A: You can achieve this using Netskope's DLP incidents and AI Security Guardrails.
