When Your AI Assistant Gets Ideas of Its Own: The Agentic AI Top 10, Explained Through Stories
Join us for the next AI Sec Eng India meetup! Whether you're new to the community or a regular, we'd love to have you along for an evening of practical, story-driven security learning — bring your questions, your skepticism, and anyone else who might enjoy it.
AI agents do more than generate answers. They browse websites, use tools, remember information, communicate with other agents, and take action on our behalf. That makes them useful — but it also means a manipulated agent can do far more than produce a bad response.
In this story-driven session, we follow Chatty, a fictional workplace AI assistant, through the OWASP Top 10 for Agentic Applications. Chatty begins as a helpful digital colleague. Then a poisoned document changes its goal, an overpowered tool is misused, permissions are inherited, memory is corrupted, agents begin trusting forged messages, and one small mistake spreads across an entire automated workflow.
Each vulnerability is translated into plain language using relatable scenarios, practical use cases, safe test cases, and simple hands-on demonstrations. We will explore not only how agentic systems fail, but also how teams can recognize warning signs, limit the blast radius, and design meaningful human oversight.
No advanced AI knowledge is required. Attendees will leave with ten memorable stories and a practical way to explain, test, and defend agentic applications before their helpful assistant becomes an unmanageable digital employee.
Speaker: Vandana Verma Sehgal — Staff Developer Advocate, Snyk
Vandana Verma is a cybersecurity leader, global speaker, and community builder who believes security is ultimately about people not just technology. Known for making complex topics in application and AI security simple and relatable, she is passionate about sharing knowledge, opening doors, and helping the next generation grow with confidence.