Overview
Skills
Job Details
Title: AI Security Engineer
Location: Remote across USA
Department: AI Security Engineering
Reports To: Head of Security Engineering
Role Overview
The AI Security Engineer designs, evaluates, and implements secure architectures for Large Language Model (LLM) and Agentic AI ecosystems across the enterprise. This includes securing platforms like ChatGPT Enterprise, Claude Enterprise, Gemini Enterprise, Google AI/Vertex AI/LM Notebooks, Azure OpenAI, Azure AI Foundry, and Model Context Protocol (MCP) environments. The role ensures robust data protection, model governance, runtime security, alignment, and compliance, bridging security architecture, AI engineering, legal compliance, and risk governance.
Key Responsibilities
AI Security Engineering & Design
- Engineer secure environments for enterprise LLM platforms (ChatGPT, Claude, Gemini, Azure OpenAI).
- Design zero-trust architectures for AI ecosystems, including MCP servers/clients and agentic workflows.
- Secure LLM model lifecycle: training, fine-tuning, evaluation, deployment, inference endpoints.
- Define agent-to-agent (A2A) trust boundaries, cryptographic trust chains, message integrity controls.
- Establish guardrails for Retrieval-Augmented Generation (RAG), tool use, plugins, function calling, enterprise embeddings, contextual memory.
- Implement runtime sandboxing, prompt firewalling, data path isolation, interaction filtering.
AI Risk, Governance & Compliance
- Apply frameworks: NIST AI RMF, MAESTRO, OWASP Top 10 for LLM & Agentic AI, MITRE ATLAS, ISO/IEC 23894 & 42001, Google SAIF, Microsoft Responsible AI Standard.
- Establish model governance, evaluation criteria, audit logs, chain-of-thought protection, policy configuration.
AI Security Threat Modeling & Controls
- Conduct threat modeling using: LLM-specific, Agentic AI Self-Propagation & Tool Abuse, RAG Architecture Security, A2A Trust Exploitation, MCP Supply-Chain & Man-in-the-Middle models.
- Define adversarial defenses: prompt injection mitigation, jailbreak prevention, indirect prompt poisoning, model exfiltration protection, data poisoning countermeasures, model inversion & membership inference prevention.
Platform Security
- Design secure Azure OpenAI & Azure AI Foundry deployments: private endpoints, VNet isolation, mTLS/encryption, model filtering, enterprise data security.
- Secure Gemini Enterprise & Google LM Notebooks: VPC Service Controls, IAM conditional access, DLP, context filtering, confidential computing.
Agentic AI & MCP Security
- Govern MCP tools, input/output sanitization, policy-guarded capability authorization.
- Define secure orchestration and oversight for multi-agent LLM systems: autonomy limits, escalation rules, tool use governance.
Model Training Security & Supply Chain Integrity
- Implement Secure MLOps: dataset lineage, provenance, quality checks, differential privacy, secure gradient computation, adversarial training, signed/documented model artifacts.
- Secure confidential training data, prevent leakage to public models.
AI Monitoring & Incident Response
- Enable runtime protection, anomaly detection, exploit signal monitoring.
- Build AI-specific incident playbooks: hallucination incidents, governance policy drift, unauthorized agent actions, emergent harmful behavior.
Required Technical Skills
6 10 years in cybersecurity, including 2+ years in AI/ML security or LLM platform engineering.
Core AI Security Expertise
- Deep understanding of generative AI security: LLM jailbreak defense, guardrails engineering, AI alignment, content filtering, advanced prompt-level security.
- Knowledge of LLM tool ecosystems (functions, plugins, RAG).
Enterprise AI Platforms
- Security configurations for ChatGPT Enterprise, Claude Enterprise, Gemini Enterprise, Google LM Notebooks, OpenAI on Azure, Azure AI Foundry.
Cybersecurity & Cloud Architecture
- Zero-trust architectures, KMS/HSM/secrets management, API/function calling security, encryption controls, network/IAM/private routing, DSPM, CASB, CSPM, AIRS tools.
Programming & Tooling (Preferred)
- Python, TypeScript/Node.js, Terraform/IaC for secure AI deployments.
- Agentic AI frameworks: LangChain, LangGraph, OpenAI Agents, CrewAI, AutoGen. ADK
AI Security Tooling Hands-On Skills & Experience
AI Runtime Security & Agent Guardrails
- OpenAI Security Capabilities, Anthropic Claude Admin APIs, Google SAIF Controls, Vertex AI Guardrails, Azure AI Foundry Governance.
- Content Filtering/Toxicity Classifiers: OpenAI Risk Filters, Perspective API, Azure Content Safety.
- Prompt Firewalls/Guardrails Engines: Prompt Armor, Guardrails AI, Prompt Shield, NeMo Guardrails.
- AI Agent Monitoring: Protect AI, Lakera, Arthur.ai, Robust Intelligence, CalypsoAI.
LLM Supply Chain Security / Secure MLOps
- Model artifact signing/integrity: Sigstore, in-toto, SLSA compliance.
- Dataset provenance: BastionML, Cleanlab, Alectio.
- Adversarial Training/Validation: IBM ART, CleverHans, TextAttack, ShieldGemma.
- Model Watermarking/Exfiltration Prevention: Watermark-LM, RIME, DeepMind SynthID.
- Pipeline enforcement: Kubeflow, Azure ML, Vertex AI Pipelines, MLflow.
Agentic AI Security & MCP Ecosystem
- MCP secure configuration tooling, policy enforcement, signed client tools.
- Secure tool API integration, capability authorization, dynamic context redaction, scope-limited tool exposure.
- Agentic AI orchestration hardening: LangGraph, OpenAI Agents, AutoGen Studio, CrewAI.
- A2A Trust Models: mTLS, token-based capability scoping, replay-attack defense, real-time behavior anomaly analytics.
Cloud Platform AI Security Tooling
- Azure: Microsoft Purview, Defender for AI, Synapse secure RAG vector DB controls.
- Google Cloud: VPC-SC, Confidential Space/Computing, DLP API, IAM ReBAC.
Threat Intel & Offensive Security Tools for LLM
- LLM Pentesting: PentestGPT, LLM-Guard, Azure AI Red Team Tools.
- Prompt Injection Scanners: PIA, picoGPT Security Test Kit.
- Model behavior fuzzing: GARAK.
- Membership inference/property leakage evaluation: PrivacyRaven