Security Engineer – Agentic AI

This is a fully remote position, open to applicants in United States.

📋 Description

• Conduct threat modeling for agents, tools, MCP servers, and multi-agent workflows utilizing OWASP Top 10 for LLM Applications, OWASP Agentic AI threats, and MITRE ATLAS.

• Implement a streamlined pre-launch security review process.

• Evaluate agentic code and designs, publish secure defaults, and clarify risk trade-offs that are explicitly owned.

• Educate on secure agent practices through manuals, threat-model walkthroughs, demonstrations, and office hours.

• Develop authentication and authorization for agents and tools with OAuth 2.0/OIDC, token exchange, scoped short-lived credentials, per-tool least privilege, and fail-closed MCP authorization.

• Strengthen containers and sandboxes through filesystem isolation, egress controls, and requiring human approval for irreversible actions.

• Create gateway defenses against prompt injection, tool misuse, excessive agency, and data exfiltration using input/output filtering, PII/DLP, and policy-as-code.

• Develop automated adversarial tests for injection, tool misuse, exfiltration, and goal drift in CI and according to a schedule.

• Manage MCP servers, skills, plugins, models, and dependencies through reviews, allow-listing, signing, SBOMs, and scanning.

• Ensure traceability of model, tool, and agent calls, while integrating guardrail events and anomalies into monitoring and alerting systems.

• Oversee controls, optimize guardrails, and maintain them updated and patched in production.

• Lead agent-related incident response and drive technical resolutions.

• Align controls with ISO 27001, ISO/IEC 42001, SOC 2, GDPR, and TPN, ensuring evidence is audit-ready.

• Create controls, tools, and tests; assess code and designs; and disseminate security knowledge across Brahma teams.

• Stay abreast of emerging agent and AI threats, evaluate new attack categories in controlled experiments, and implement proven defenses into production.

⛳️ Requirements

• Over 5 years of experience in security engineering with a significant focus on application or product security.

• Proficient in Python.

• Proven experience in delivering security tools, controls, and automated tests as production-ready code.

• Practical experience in securing LLM applications or agents against prompt injection and unsafe tool usage.

• Extensive knowledge of OAuth 2.0, OIDC, token exchange, workload identity, and least-privilege design principles.

• Experience in leading structured threat modeling efforts and translating findings into prioritized engineering tasks.

• Familiarity with integrating security scanning and checks into CI/CD processes.

• Experience in securing containerized workloads.

• Ability to clearly articulate risks and trade-offs to engineers and leadership, along with thorough documentation of decisions.

• A history of educating and documenting security practices for peers and other teams.

• Nice to have: experience with garak, PyRIT, promptfoo, or similar tools.

• Nice to have: hands-on experience with OPA/Rego, Cedar, or similar technologies.

• Nice to have: knowledge of gVisor, Firecracker, seccomp, or Landlock.

• Nice to have: experience in building or securing MCP servers, LLM gateways, or guardrail services.

• Nice to have: working knowledge of GCP IAM, Workload Identity, Secret Manager, and Cloud Logging.

• Nice to have: exposure to ISO/IEC 42001, NIST AI RMF, or MPA/TPN content security frameworks.

• Nice to have: OSCP, GIAC, CSSLP, or equivalent certifications.

🏝️ Benefits

• No formal on-call rotation currently, although one may be implemented in the future.