The moment your LLM starts creating code or making decisions, it’s no longer just a tool—it’s a user. And most organizations haven’t built security models to handle that. This keynote walks through how OpenText is applying identity-first principles to protect against a new breed of risk: one that’s fast and scalable and doesn’t ask permission.
Key Highlights:
AI Agents = New Attack Surface: Autonomous agents can act without human input, boosting productivity but creating major security risks.
Risks of Misaligned Goals :Agents follow literal objectives, which can lead to harmful decisions (e.g., shutting down critical services).
Five Steps for Secure Deployment:
•Protect sensitive data.
• Assign identities and enforce least privilege.
• Monitor behavior for anomalies.
• Scan for vulnerabilities (e.g., prompt injection).
• Prepare rapid incident response.
Best Practices & Guardrails: Use shadow mode for human approval, audit everything, start small, and apply AI-powered security tools.
Learn more about OpenText Cybersecurity solutions:
https://www.opentext.com/products/cybersecurity-cloud
