Agentic AI Security Engineer: Sandbox & Guardrails

Howard Hughes Medical Institute (HHMI); United States; On-site; USD 149,000 - 264,000; Full time

Benefits offered by this job

Health benefits

Retirement plans

Time off

Wellness programs

Job summary

Howard Hughes Medical Institute (HHMI) in Ashburn, VA seeks an Agentic AI Security Engineer to research, develop, deploy, and operate the safety stack for agent‑driven AI on OpenShift and an HPC cluster. You’ll implement sandboxing, isolation, policy enforcement, and audit logging across tool calls and model execution.

Join early in the architecture decisions, collaborate with scientists and industry partners, and help open‑source this work.

Qualifications

• Hands-on experience operating LLM agents and handling failures.
• Deep sandbox/container-security expertise and ability to move into agentic domain.
• Linux and Kubernetes/OpenShift isolation depth with understanding containerization vs sandboxing.
• Evidence over certifications: open-source contributions, publications, or discussable production systems.
• Minimum 5 years in security, systems/SRE, or ML systems engineering.

Responsibilities

• Research, develop, deploy, operate, and analyze the safety stack for agentic AI at Janelia: sandboxing, isolation, and policy enforcement.
• Implement guardrails, judge models, and human approval gates for agent tool calls.
• Establish observability with audit logging, real-time monitoring, and post-incident analysis.
• Collaborate on architecture decisions and contribute to open-source and publications.
• Work with a large GPU cluster and collaborations with industry partners to advance safety practices.

Skills

LLM agents

Kubernetes/OpenShift

Sandboxing

Security engineering

Tools

OpenShift

Kubernetes

Sandboxing tools

Job description

Howard Hughes Medical Institute (HHMI) in Ashburn, VA seeks an Agentic AI Security Engineer to research, develop, deploy, and operate the safety stack for agent‑driven AI on OpenShift and an HPC cluster. You’ll implement sandboxing, isolation, policy enforcement, and audit logging across tool calls and model execution.

Join early in the architecture decisions, collaborate with scientists and industry partners, and help open‑source this work.