The Agentic Shift: How to Empower AI Without Losing Control

Session Info: When Agents Go Rogue: Securing Autonomous AI Before It Secures Your Breach AI agents are rapidly evolving from passive assistants into autonomous systems that can reason, plan, use tools, access enterprise data, and execute workflows with limited human supervision. That shift creates a new security problem: the risk is no longer just unsafe model output, but unsafe action. Recent incidents and research have shown AI agents being hijacked through prompt injection, misused through excessive tool permissions, manipulated through untrusted integrations, and leveraged in cyberattacks that automate reconnaissance, exploitation, lateral movement, and data exfiltration. This presentation examines how the threat model changes when AI systems can act, not just answer. It covers the foundations of agentic systems, how agents interact with tools and with each other, why identity and authorization are critical for safe deployment, and what recent cyberattacks involving AI agents reveal about the evolving nature of offensive operations. The session concludes with practical design principles for securing AI agents through bounded autonomy, explicit trust boundaries, strong policy enforcement, and runtime monitoring. The goal is to give the audience a clear mental model for defending agentic systems before autonomy becomes an attacker advantage. Presented by Vignishwari:   / vigneshwari-v-b3457125   Hosted By ‪@SaaviGenAI‬ empowers organizations and working professionals to succeed in the AI era through AI Security, hands-on upskilling, and AI-powered cybersecurity solutions. We help professionals build practical skills in Generative AI, Agentic AI, LLM Applications, RAG, and AI Security while helping businesses secure and govern AI adoption. Our flagship solutions, SaaviSOC and SaaviAIDR, provide AI-driven Security Operations, Threat Detection, Incident Response, and Cyber Defense capabilities designed for modern enterprises. Nanda Kumar is a cybersecurity leader and AI practitioner with more than 23 years of experience spanning product engineering, security operations, incident response, compliance, and enterprise technology leadership. Throughout his career at Cisco, HPE Aruba, and ChargePoint, he has worked across both offensive and defensive security domains, helping organizations build secure products, operate resilient infrastructures, and scale security programs globally. At Cisco, Nanda contributed to the development of the CS-MARS SIEM platform and Cisco's Next-Generation Firewall technologies. He also trained global technical and customer-facing teams on Cisco's security portfolio, helping drive adoption and operational excellence across worldwide deployments. At HPE Aruba, he contributed to the IntroSpect User and Entity Behavior Analytics (UEBA) platform and later led Cybersecurity Incident Response (CSIRT) initiatives, while driving security and compliance programs including SOC 2 Type II, ISO 27001, GDPR, and enterprise governance requirements supporting Fortune 100 customers. His experience across security product development, threat detection, incident response, governance, and large-scale enterprise operations has shaped a practical philosophy: design for detection, instrument for forensics, and operate securely at scale. Today, Nanda focuses on Applied AI, LLM Security, and secure AI adoption. He advises organizations on building, deploying, and governing AI systems responsibly while helping development teams, product leaders, security professionals, and business executives leverage AI effectively in real-world environments. In addition to consulting and advisory work, he trains professionals across Fortune 100 organizations on Applied LLMs, AI-driven workflows, secure AI engineering, and practical enterprise adoption strategies. LinkedIn:   / nandakumar80   #ai #saavigenai #saaviSOC #saaviAI #claude #cybersecurity #aisoc #aisecurity