AI Security Architect

We are seeking an experienced and highly skilled AI Security hands-on, highly technical architect responsible for defining security architecture and implementing robust security controls for our AI/ML systems and their underlying platforms and will serve as the team’s technical mentor and architecture authority, driving secure-by-design patterns across the AI/ML lifecycle (data, training, evaluation, deployment, and production monitoring) and proactively mitigating AI-specific threats such as model integrity risks, data poisoning, adversarial attacks, prompt injection, model extraction, and inference-time abuse. Lead technically, set standards, and guide engineers day-to-day through architecture, reviews, and delivery.

Ensures AI systems are secure, compliant, and resilient by implementing data protection, threat detection, guardrails, and ongoing risk monitoring across the AI lifecycle. Platform & Enablement Roles AI Platform Admin (M365, copilot Studio) Manages AI platforms and environments, including access provisioning, governance controls, and policy enforcement (e.g., DLP, security, and compliance). AI Reusable Utility Develops reusable components (e.g., prompts, connectors, APIs, templates) to accelerate AI solution delivery and promote standardization across use cases. AI Common Infrastructure, Framework & Observability Architect (AWS and Azure) Designs and maintains the foundational AI infrastructure, frameworks, and observability capabilities (telemetry, monitoring, metrics) required for scalable, reliable, and governed AI operations.

Core Responsibilities Agent Security Non-Human Identity & Access: Define strict Role-Based Access Control (RBAC) and least-privilege models for AI agents using identity systems (e.g., Entra Agent ID). Guardrails & Sandboxing: Design runtime environments with restricted permissions to prevent manipulated agents from accessing unauthorized APIs, data sources, or executing malicious toolchains. Input/Output Protection: Implement defenses against adversarial attacks, prompt injections, jailbreaking, and sensitive data leakage (DLP) across agent workflows. Observability & Monitoring Decision Traceability: Architect logging and monitoring standards to map how reasoning agents use data and call APIs, eliminating “black box” decisions. Model Drift & Integrity: Monitor models and prompt templates in production to detect behavioral drift, anomalies, and poisoning or evasion attacks. SOC Monitoring & Automation Autonomous Security (AI SOC): Design LLM-driven and agentic workflows to improve alert triage, contextual correlation, false-positive filtering, and playbook automation. Incident Response Playbooks: Establish remediation strategies and threat-hunting procedures for AI-specific events (e.g., compromised model artifacts, hallucination-driven exploits). Compliance Enablement & Governance Regulatory Alignment: Map AI-specific controls to established standards like the NIST AI RMF, OWASP Top 10 for LLMs, and GDPR. Audit Readiness: Build audit pipelines that track and explain everything an agent does to satisfy ongoing AI regulatory compliance and governance requirements. Architecture & Secure-by-Design Leadership Define and maintain AI security reference architectures for multiple AI deployment patterns, including MCP / Agentic AI and LLM application stacks (RAG, tools/plugins, agents, orchestration). Establish and evolve security requirements, patterns, and guardrails across the AI/ML SDLC (design * build * run), including secure pipelines and platform controls. Own AI security architecture decisions across critical domains: identity, secrets, data protection, network controls, tenancy boundaries, logging/telemetry, and isolation for training/inference. Control Design & Implementation (Hands-on) Design and deploy controls to ensure model integrity and governance, including RBAC/ABAC for models, feature stores, data sets, registries, and evaluation artifacts. Build/enable technical mechanisms for provenance, attestation, signing, and approval workflows (where applicable) across datasets, models, prompts, and deployments. Drive implementation of runtime protections for AI services (abuse prevention, rate limiting, input/output validation, prompt-injection mitigations, model endpoint hardening, and monitoring). Threat Modeling, Assurance, and Risk Reduction Conduct and lead AI/ML-specific threat modeling (data poisoning, model evasion, extraction, inversion, supply-chain, prompt attacks), translate findings into actionable backlogs, and drive remediation. Define and run security design reviews for AI initiatives; provide clear, pragmatic architecture guidance and document exceptions with risk acceptance paths. Establish AI security testing approaches (adversarial testing, red-teaming enablement, evaluation security, misuse/abuse cases) and integrate into delivery pipelines. Tooling, Automation, and Operational Enablement Design and deliver AI security tooling to improve and automate cybersecurity posture (e.g., controls coverage, policy-as-code, detection engineering, vulnerability management integration, incident response playbooks for AI-specific events). Define logging/monitoring standards and detection use-cases for AI platforms and LLM apps (drift signals, anomalous access, suspicious prompt patterns, exfiltration indicators, policy violations). Technical Mentorship & Influence (No Line Management) Act as the team’s technical mentor: coach engineers through designs, implementations, and trade-offs; raise engineering quality via reviews, pairing, and knowledge sharing. Lead by influence across Data Science, Engineering, Product, Platform, and Cybersecurity-driving alignment without formal authority. Create internal enablement materials: runbooks, architecture standards, reusable patterns, and reference implementations.

Experience: 7+ years in cybersecurity architecture with proven experience securing large-scale LLM deployments and multi-agent workflows. Technical Proficiency: 5+ years of hands-on capability with agent frameworks (e.g., LangChain, LangGraph, AutoGen) and MLOps platforms. Framework Knowledge: 3 to 5 years of Deep familiarity with model risk management principles and AI security standards

Common Expectation from all the roles: Compliance with Client’s responsible AI principles and Acceptable Use policy Adherence to data residency, privacy (GDPR, HIPAA where applicable), and 21 CFR Part 11 controls where in scope Third-party risk assessment and SOC 2 Type II (or equivalent) certification Disclosure of subcontractors and offshore delivery locations Disclosure of model providers, training data practices, and any use of client data for model improvement (opt-out required)