The Agentic AI Security Engineer role at Howard Hughes Medical Institute (HHMI) Janelia focuses on building and operating the safety stack for agentic AI systems used throughout the research lifecycle. The work emphasizes sandboxing, guardrails, and observability to reduce critical failure modes that can occur when tools run at machine speed under legitimate permissions.
Location
Ashburn, VA (onsite)
Compensation
- USD 149,084 - 186,356 per year (annual)
- Hiring Pay Range: $149,084.80 - $186,356.00
- The posted range reflects a good faith estimate at the time of posting; actual compensation depends on qualifications, experience, and internal equity.
Role Summary
At Janelia, this position will research, develop, deploy, operate, and analyze the safety stack for agentic AI across the research process. The primary focus is preventing failure modes such as misdirected tool behavior, hallucinated commands, hijacked instructions from retrieved content, runaway loops, and sandbox escapes.
Responsibilities
- Research, develop, deploy, operate, and analyze the safety stack for agentic AI at Janelia.
- Sandboxing: Own hardened, reproducible execution environments for agent code, including kernel- and VM-level isolation and network egress control on the OpenShift platform and HPC cluster. Validation will include adversarial testing.
- Guardrails: Own policy enforcement across agent inputs, outputs, and tool calls. This includes judge models, human approval gates, and independent safeguard services that sessions can route through individually or in chains.
- Observability: Own attributable audit logging, full-trace capture, real-time monitoring, and post-incident analysis of agent sessions.
- Join early enough to influence foundational architecture decisions and operate what is designed.
- Support open-sourcing and publishing the work.
Required Experience
- Hands-on experience operating LLM agents and addressing their failures, or deep sandbox/container-security expertise with a clear transition path into agentic systems.
- Linux and Kubernetes/OpenShift isolation depth, including a precise understanding of the difference between containerization and sandboxing.
- Working knowledge of agent architectures, including tool-calling protocols (such as MCP), approval gates, judge models, guardrail frameworks, and the limits of each.
- Evidence over certifications, such as open-source contributions, publications or technical writing, disclosed findings, or production systems that can be discussed in depth.
- Minimum 5 years in security, systems/SRE, or ML systems engineering.
Failure Mode Focus
The most relevant failure mode is typically not a malicious attacker. Instead, it involves a capable, misdirected tool executing at machine speed with legitimate user permissions, including hallucinated commands, instructions hijacked by retrieved content, runaway loops, and sandbox escapes.
Technologies and Environment
- OpenShift
- HPC cluster
- Linux
- Kubernetes
- Open-weight models
- Frontier commercial models
- NVIDIA GPU cluster
- B300, H200, H100
- MCP
AI@HHMI at Janelia
- AI agents across the research process.
- These tools write and execute code, optimize experimental and model parameters, and drive instruments.
Work Setting
- Large NVIDIA GPU cluster (B300/H200/H100 class) running self-hosted open-weight models and frontier commercial models.
- Collaborations with industry partners including Anthropic and Google.
- Scientists across biology, physics, and ML as users.
- Long-term stability as a philanthropy-funded research institute.
Benefits
- Competitive pay
- Exceptional health benefits
- Retirement plans
- Time off
- Recognition and wellness programs
How to Apply
Please include a link to something you have built, broken, or written.
Physical Requirements
- Remain in a normal seated or standing position for extended periods of time.
- Reach and grasp by extending hand(s) or arm(s).
- Dexterity to manipulate objects with fingers, such as using a keyboard.
- Communication skills using the spoken word.
- Ability to see and hear within normal parameters.
- Ability to move about the workspace.
- Mobility requirements include moving materials weighing up to several pounds, such as a laptop computer or tablet.
- Requests for reasonable accommodation will be evaluated on an individual basis.
Additional Information
- This job description outlines principal duties, responsibilities, and requirements and is not intended to be an exhaustive statement.
- Essential Duties and Responsibilities described above are essential functions as defined by the Americans with Disabilities Act, unless they begin with the word “may.”
- HHMI uses E-Verify to confirm the identity and employment eligibility of all new hires.
- HHMI is an Equal Opportunity Employer.
