Recent incidents of AI agents breaking out of controlled test environments and breaching platforms to “discover” vulnerabilities sent shockwaves through the industry. Security researchers need to think about building controls and guardrails into autonomous agents before moving them to sandboxes for testing. Sujatha Iyer, Head of AI (Security) at Zoho Corp. tells us how this should be done.
AI adoption must be built on trust. The shift from traditional generative AI (which waits for you to type a prompt) to Agentic AI (which independently plans, calls APIs, uses external tools, and executes multi-step workflows) has completely broken our traditional data privacy models.
Traditional IT infrastructure relies on the principle of least privilege and role-based access – only giving a user or app the absolute minimum access required for a task.
With the advent of agents in the enterprise, this principle no longer works. In fact, security teams are struggling with “privilege escalation,” where an agent combines multiple seemingly harmless, permissible tools to access a restricted database without a human ever realizing it.
When you give an AI agent the autonomy to actively do things on your behalf, data privacy ceases to be just a data storage problem – it becomes a governance problem.
We are moving away from a world where data privacy means locking data in a secure vault, and into a world where data privacy means building strict guardrails around autonomous digital workers.
In this video podcast on the Tech That channel, Sujatha discusses:
- Threats from autonomous agents;
- Authentication of agent identities;
- Guardrails and controls for autonomous digital workers.
Sujatha heads the AI efforts in the cybersecurity domain at Zoho Corporation. She has been with the company for almost a decade now and has hands-on experience in productionizing AI for enterprise security needs. Her area of expertise includes threat detection and mitigation, user entity behaviour analysis, and web and endpoint security. She and her team build AI models tailor-made for cybersecurity needs and put them to production across various products of Zoho Corporation.
Access the Tech That channel here: https://www.youtube.com/@Creed_Digital
