Inside the Mind of a Machine: Adversarial Tricks That Fool AI Systems

Pallab Haldar

Abstract:

AI systems now power security tools, fraud detection, identity checks, and automated decisions. Yet these systems can be confused, redirected, or manipulated in ways that are strange, creative, and sometimes hilarious. This session explores the world of adversarial manipulation where tiny inputs can redirect a model into false beliefs or dangerous actions.

The talk explains how attackers use crafted text, images, sensor noise, or misleading signals to bypass AI filters and mislead automated security controls. Attendees will learn how small prompts can slip past policy rules, how images can fool classifiers, and how timing tricks can confuse real time behavior models. The session includes live demonstrations that reveal how models respond in unexpected ways when faced with deception.

This session is designed as a bridge between AI research and the hacker mindset. It shows how curious exploration can uncover weak points and how defenders can build simple practices that reduce risk. No previous AI background is needed, only curiosity and a laptop. By the end, participants will understand how adversarial thinking can strengthen the next generation of security and privacy systems.

Pallab Haldar

Technical Architect, Cognizant Technology Solution

Pallab Haldar is an accomplished SAP HANA Data Architect and Generative AI Specialist with over 19 years of transformative experience in enterprise data architecture, advanced analytics, and AI/ML innovations. Currently serving as SAP Technical Data Architect at Cognizant Technology Solutions, he has delivered high-impact solutions for Fortune 500 clients including 3M, Delta, and Yokohama Tire.
Pallab specializes in architecting enterprise-scale SAP HANA Database solutions, seamlessly integrating data models with core business processes across SD, MM, FI, BW, and analytics domains. His expertise spans the complete data lifecycle—from ingestion and provisioning to governance and analytical consumption. As a recognized SAP GenAI expert, he pioneers AI-driven transformations, implementing conversational AI assistants with SAP Joule, predictive forecasting in SAP Analytics Cloud, and RAG pipelines using LangChain and Haystack.
His technical mastery encompasses SAP HANA performance optimization, XSA application development, and cloud platform integration across SAP BTP, Cloud Foundry, and Neo environments. Pallab has successfully led complex migrations from XSC to XSA, engineered ML models for predictive analytics, and developed deep learning solutions using TensorFlow, Keras, and PyTorch.
A thought leader in the SAP community, Pallab actively shares knowledge through technical blogs and holds 15+ SAP certifications, including SAP Generative AI Developer and multiple Solution Architect credentials. His innovative approaches to database optimization, query performance tuning, and AI/ML adoption have consistently delivered measurable business impact through process optimization and enterprise automation.
With deep expertise in programming languages including SQL, Python, Java, and ABAP, combined with extensive experience in DevOps practices and cloud technologies, Pallab continues to shape the future of enterprise data architecture and AI integration. He is an active IEEE member and maintains a strong presence in the professional community through his contributions and thought leadership.