Naga Sujitha Vummaneni

Senior Security Engineer,

Ripple

ABOUT THE SPEAKER:

I’m a Senior Security Engineer at Ripple and an Executive MBA graduate of Cornell SC Johnson College of Business, with over 10 years of experience securing large-scale cloud and blockchain infrastructure at organizations including Google, Nike, eBay, and Cisco.
My research sits at the intersection of machine learning and adversarial security — spanning game-theoretic prompt injection frameworks, zero-knowledge ML for blind signing prevention, federated threat detection, and RAG-based cybersecurity systems. I’ve published work across IEEE venues on LLM security, neuromorphic computing, and decentralized trust models.
At Ripple, I lead security engineering across multi-cloud environments (AWS, Azure, GCP), applying ML-driven automation to threat detection, incident response, and compliance at blockchain payments scale. My practitioner perspective bridges the gap between theoretical ML security research and the operational realities of deploying AI in high-stakes financial infrastructure.
I hold AWS Certified Security Specialty and CISM certifications, serve as an Associate Fund Manager with Big Red Ventures at Cornell, and am an active TPC reviewer for IEEE conferences. I speak and write on the security implications of decentralization — including the tension between trustless systems and centralized control vectors that make blockchain environments uniquely vulnerable.

TALK TITLE:

Jailbreaking the Blockchain: How I Used Game Theory to Map Prompt Injection Attack Surfaces in Agentic Systems

TRACK:

Technical / Engineering Talks

SUB TOPIC:

Fine-Tuning & Training – Safety / Governance / Auditability

ABSTRACT:

Most AI agent evaluation frameworks are built to measure capability, not adversarial robustness. The result: agents that ace benchmarks but collapse under real-world attack conditions — prompt injection, goal hijacking, tool misuse, and output trust exploitation that standard evals never surface.

Drawing on research developing game-theoretic prompt injection frameworks and zero-knowledge ML systems for high-stakes financial infrastructure at Ripple, this session presents a concrete methodology for adversarial agent evaluation that practitioners can apply to their own pipelines.

You’ll leave with a working model for mapping your agent’s attack surface across orchestration logic, tool boundaries, and downstream trust chains — and a set of design patterns for building agents that are auditable and adversarially robust by architecture, not by accident. Whether you’re building coding agents, deploying LLM workflows in production, or responsible for governing AI systems at scale, this session reframes how you think about evaluation before your next deployment.

WHAT YOU’LL LEARN:

Agent vulnerability is primarily architectural, not a model alignment problem — fixing the model without addressing orchestration logic leaves the most exploitable attack surface untouched

Standard eval suites systematically miss adversarial failure modes because they don’t model an active adversary — game-theoretic stress testing catches what benchmarks don’t
The four-layer attack surface framework (input, tool use, inter-agent trust, output) gives practitioners a checklist-level structure for auditing any agent deployment
Zero-knowledge commitments are a viable primitive for agent auditability at scale — implementable with acceptable latency tradeoff for high-stakes pipelines
Evaluation methodology designed for adversarial robustness doubles as a governance artifact — teams facing compliance scrutiny on AI agent deployments can use the same framework to answer “how was this tested”
Partial architectural fixes (e.g., input sanitization only) can create false confidence — the game-theoretic model shows how attackers route around single-layer defenses

Naga Sujitha Vummaneni

Who Attends

2023 Event Demographics

2023 Technical Background

2023 Attendees & Thought Leadership