Senior Lead Security Architect, AI/ML Platforms

What you'll do

• Design secure architectures for AI applications, agentic workflows, and machine learning systems to ensure adversarial resilience.
• Conduct threat modeling and adversarial testing on generative AI, RAG pipelines, and ML systems to identify vulnerabilities like prompt injection.
• Develop and implement security controls across the full model development lifecycle (MDLC) and agent runtime environments.
• Define authentication, authorization, and least-privilege access controls for users, services, and tools within AI ecosystems.
• Establish logging, monitoring, and observability standards to ensure auditability of prompts, tool calls, and model outputs.
• Evaluate orchestration patterns to enforce safe execution boundaries and human-in-the-loop requirements for automated agents.
• Secure integration patterns for Model Context Protocol (MCP) and other tool-connection mechanisms to prevent data exfiltration.
• Translate emerging AI safety research and industry standards into actionable engineering guardrails and technical policies.

What we're looking for

• 5 years of applied experience in cybersecurity architecture and/or securing AI/ML systems.
• Practical cloud-native experience in AWS, GCP, or Azure using Public Cloud AI/ML services like SageMaker or Bedrock.
• Advanced proficiency in one or more programming languages to review code and architecture for security.
• Advanced knowledge of cybersecurity architecture and in-depth expertise in artificial intelligence and machine learning.
• Experience with ML development workflows including Python, TensorFlow, PyTorch, and the model development lifecycle (MDLC).
• Solid understanding of AI system attack surfaces, including prompt injection, data poisoning, and retrieval risks.
• Working knowledge of AI agent security fundamentals such as authN/authZ, secure tool use, and logging for audit.
• Experience with API security, container orchestration, IaC, or relevant certifications (preferred); familiarity with MCP standards (preferred).