AI Security SME
Posted
Description
You will lead the security architecture, governance, and assurance of our AI and GenAI platforms across the organization. This includes classical ML systems and modern LLM/agentic architectures. You will define how models are designed, deployed, monitored, and defended; ensure they are robust, explainable, privacy-preserving, and compliant; and act as the go-to expert for AI security across products, data, cloud, and security teams.
- AI Security Architecture
- Design and review secure architectures for ML/LLM workloads (training, fine tuning, inference, RAG, agents, plugins, tool calling, APIs).
- Define reference architectures for on-prem, hybrid, and cloud AI platforms (Azure OpenAI, AWS Bedrock, GCP Vertex, self-hosted models, etc.).
- Threat Modeling & Risk Management
- Perform AI-specific threat modeling (e.g., data poisoning, model theft, prompt injection, jailbreaks, supply-chain, inference attacks) using MAESTRO or similar framework.
- Align controls with leading frameworks: NIST AI RMF, ISO/IEC 27001, ISO/IEC 27090, ISO/IEC 42001, OWASP GenAI / LLM Top 10, CSA & MITRE ATLAS.
- Security Control Design & Implementation
- Define and oversee implementation of controls for:
- Training & data pipelines (data quality, provenance, labeling, PII protection).
- Model & artifact integrity (signing, SBOM, secure registries).
- Access control, isolation, rate limiting, and abuse detection.
- Secure prompt engineering and guardrail policies.
- Security prompt monitoring to detect ongoing attacks.
- Data security: data classification, data masking and DLP.
- MCP server hardening.
- Define and oversee implementation of controls for:
- Reliability & Trustworthiness
- Partner with engineering and data science to embed robustness, observability, fallback strategies, and evaluation pipelines (safety, bias, toxicity, hallucination monitoring).
- Contribute to SLOs/SLAs for AI systems, including security and reliability KPIs.
- Secure SDLC for AI
- Embed AI security into CI/CD: scanning, dependency checks, policy-as code, red-teaming AI components pre- and post-release.
- Incident Response & Red Teaming
- LLM bias detection.
- Perform red teaming activities to abuse and force AI hallucinations.
- Develop and maintain AI-specific playbooks (prompt abuse, model exfiltration, data leakage, compromised agents).
- Lead or support AI red/blue/purple teaming exercises using frameworks like MITRE ATLAS.
- Governance, Compliance & Policy
- Advice on alignment with emerging AI regulations and standards (e.g., EU AI Act, regional laws, internal AI use policies).
- Define internal policies on responsible AI, data usage, model lifecycle, and 3rd-party AI risk management.
- Stakeholder Leadership
- Run workshops, training, and awareness for engineering, security, and business teams.
Required Qualifications & Experience:
- 8–12+ years in Cybersecurity, with 3–5+ years focused on AI/ML or data platforms (can be overlapping).
- Hands-on experience with:
- Cloud platforms (Azure, AWS, GCP) and their AI services.
- At least one Agentic/GenAI stack (e.g., Transformers, LangChain/LlamaIndex, vector DBs, model gateways, MLOps platforms).
- Proven track record designing or reviewing secure architectures for:
- ML pipelines, LLM/RAG systems, or agentic/automation platforms.
- Strong understanding of:
- Cryptography, identity & access management, network & app security.
- Data protection & privacy (PII, PHI, DPIA concepts).
- Experience working with or mapping to frameworks/standards such as:
- NIST AI RMF, ISO/IEC 27001, ISO/IEC 42001, SOC 2, OWASP Top 10 & OWASP GenAI/LLM Top 10, MITRE ATT&CK/ATLAS, CSA guidance.
- Excellent communication skills: able to translate complex AI risks into clear business and technical requirements.
Core Technical & Domain Skills:
- AI/ML & GenAI fundamentals:
- Model types (LLMs, encoders, diffusion, classical ML), training/inference flows.
- Data pipelines, feature stores, embeddings, vector databases.
- AI Penetration testing:
- Prompt injection, model tampering, data poisoning, output manipulation, exfiltration, shadow AI, insecure plugins/integrations.
- Platform & Infra:
- Kubernetes, containers, API gateways, secrets management, zero trust.
- Observability:
- Logging, tracing, safety & quality metrics for AI systems.
- Health and latency monitoring.
- Strong scripting/automation (Python preferred) for security tooling and assessments.
Certifications & Training (Required / Highly Desirable):
Essential:
- One or more core security certifications:
- CISSP, CISM, CCSP, ISO 27001 Lead Implementer/Lead Auditor.
- One or more cloud security certifications:
- CCSK (Cloud Security Alliance), AWS Security Specialty, Azure Security Engineer, Google Professional Cloud Security Engineer.
- AI & AI Security–specific training / certs (or commitment to obtain within 6–12 months):
- NIST AI RMF or ISO/IEC 42001-focused training.
- Certified AI Security Professional (CAISP / similar offerings that cover LLM/GenAI threats, MITRE ATLAS, OWASP LLM Top 10).
- OffSec / similar LLM & AI Red Teaming or GenAI security courses.
- Vendor AI/ML certifications (Azure AI Engineer, AWS Machine Learning Specialty, GCP ML Engineer) with demonstrated security emphasis.
Additional desirable:
- GIAC/GWAPT/GXPN/GCLD or similar offensive / cloud / Appsec certs.
- Formal training in:
- Secure MLOps & ML supply-chain security.
- Privacy engineering & data protection (e.g., CIPT, CDPSE).
- OWASP GenAI / LLM Top 10, MITRE ATLAS, CSA MAESTRO & other AI risk frameworks.
Skills
- Agentic AI
- AI
- API
- Application Security
- Automation
- AWS
- AWS Bedrock
- Azure
- CI/CD
- Cism
- Cissp
- Cloud
- Cloud Security
- Cryptography
- Cybersecurity
- Data Pipelines
- Data Quality
- Data Science
- DLP
- Embeddings
- Fine Tuning
- GCP
- Generative AI
- ISO 27001
- Kubernetes
- LangChain
- LlamaIndex
- LLM
- Machine Learning
- MCP
- MLOps
- Nist
- Observability
- OpenAI
- OWASP
- Penetration Testing
- Prompt Engineering
- Python
- SBOM
- SDLC
- Secrets Management
- SOC 2
- Threat Modeling
- Transformers
- Vector Databases
- Zero Trust
Sign in to read and join the discussion.