• Own the security strategy for frontier model access and MCP governance. Define how we securely connect to frontier models and external tool integrations: the risk framework, the authorization architecture, and the escalation model before those patterns harden at scale. Set the direction that builders implement within • Architect the identity and trust model for non-human agents. Define how agents authenticate, how trust is established and revoked across orchestration chains, and how the model evolves as agent capabilities expand, covering service identities, scoped credentials, and least-privilege access patterns across the platform. • Set the adversarial defense posture for AI systems in production. Define our approach to prompt injection defense, adversarial input handling, and behavioral monitoring patterns and establish the detection philosophy, telemetry requirements, and response framework the team builds and operates within. • Shape security architecture for the common AI end-user platform. Lead design reviews and build the access controls, data boundary enforcement, and abuse detection that keep a shared AI environment safe across an employee population with varying privilege levels. • Secure the shared knowledge layer. Define access control and data governance for retrieval and reasoning, ensuring AI-powered tools don't inadvertently surface sensitive data to the wrong systems or users. • Build AI supply chain integrity into the platform. Develop model provenance practices, service vetting, and dependency controls that keep the AI stack trustworthy as it grows. • Partner with Privacy, Legal, and Data Platform to ensure the right controls are built into pipelines handling real-time location, family relationship data, and data involving minors.

• Develop and answer open-ended AI security research questions that helps NVIDIA understand, measure, and reduce risk in frontier models, agentic systems, AI platforms, and AI-enabled products. • Develop practical methods, prototypes, evaluations, or tools that reveal how AI systems can fail under adversarial conditions and how those risks can be mitigated. • Explore a range of AI security problems, such as LLM and agent security, adversarial testing, model evaluation, cyber-defense automation, vulnerability discovery, secure deployment, or autonomous response. • Translate research into usable outcomes for engineering and security teams, including proof-of-concept demonstrations, benchmarks, technical guidance, mitigations, and secure-by-design recommendations. • Collaborate across offensive security, product security, AI research, platform, cloud, and infrastructure teams to connect research insights with NVIDIA's highest-impact security priorities. • Help shape NVIDIA's AI-security research strategy by mentoring others, identifying emerging risks, and building repeatable practices for evaluating and defending AI systems.

• Lead the development of security architecture guidance and standards • Assess and evaluate security postures and identify protection needs • Develop and deliver reports and strategic enterprise solutions • Support risk management and compliance activities

• Own AI behavior monitoring: Define what trustworthy and untrustworthy AI behavior looks like, and ensure it is measurable in production. • Own AI observability standards: Establish telemetry, tracing, logging, and alerting requirements for AI systems and agentic workflows. • Own control validation for agentic systems: Verify that guardrails, policy checks, access boundaries, and execution constraints are functioning as intended. • Own AI security event analysis: Detect, investigate, and document suspicious, unsafe, or non-compliant AI behaviors and coordinate response. • Own implementation support for governance frameworks: Translate governance principles into technical and operational requirements that product and platform teams can adopt. • Own AI trust metrics and reporting: Define KPIs, KRIs, and dashboards that show leadership whether AI systems are operating within approved trust and security boundaries. • Own continuous improvement of AI controls: Use incidents, testing, behavioral findings, and stakeholder feedback to strengthen control design and reduce residual risk over time.