Senior AI Security Engineer

Fatima Group; Lahore; Presencial; PKR 3 000 000 - 6 000 000; Tempo integral

Resumo da oferta

Fatima Group is seeking a software engineer with hands-on ML/LLM security experience to protect AI products from cyber threats. You will analyze and remediate findings from penetration tests and AI red-team exercises across code, models, pipelines, and cloud configurations.

You will build guardrails around model endpoints, secure patterns for RAG and agents, and enforce security checks in CI/CD. You will also contribute to client engagements where remediation is in scope and mentor peers on

Qualificações

• 5–7 years in software engineering or security, with hands‑on LLM/ML in production.

Responsabilidades

• Triage penetration test and AI red‑team findings, reproduce root causes and fix them.
• Ship fixes across code, APIs, model serving layers, data pipelines and cloud infra.
• Add regression tests including adversarial prompt tests in CI and report remediation status to leads/clients.
• Design and deploy guardrails around model endpoints and RAG/agent patterns.
• Secure ML supply chain: signing, scanning and provenance of model weights/datasets.
• Train developers on secure AI patterns and publish reusable libraries/templates.

Conhecimentos

LLM engineering

AI threat mitigation

Guardrails & design

App security fundamentals

ML supply chain

DevSecOps & CI/CD

Containers & cloud security

Formação académica

Bachelor's/Master's in CS/Security

Ferramentas

LangChain

LlamaIndex

NeMo Guardrails

Llama Guard

Lakera

Presidio

ModelScan

Garak

PyRIT

Promptfoo

Descrição da oferta de emprego

Job Description

We are looking for a software engineer first, with hands‑on ML/LLM experience from a security perspective. You take findings from penetration tests and AI red‑team exercises, find the root cause in code, models, pipelines or cloud config, and ship durable fixes.

Beyond fixing, you build the defences that stop the same class of issue returning: guardrails around model endpoints, secure patterns for RAG and agents, and security checks in CI/CD. You work on our own AI products and support client engagements where remediation is in scope.

Key Responsibilities

Remediation:

• Triage penetration test and AI red‑team findings, reproduce them and identify root cause.
• Ship fixes in application code, APIs, model‑serving layers, data pipelines and cloud infrastructure.
• Add regression tests for every fixed issue, including adversarial prompt tests in CI. Track findings to closure and report remediation status to engineering leads and clients.

AI/LLM defences

• Design and deploy guardrails around model endpoints: input and output filtering, prompt‑injection detection, PII redaction and policy enforcement.
• Build secure patterns for RAG and agents: least‑privilege tool access, sandboxed execution, human approval for risky actions, and tenant isolation.
• Secure the ML supply chain: model weights, datasets and dependencies (signing, scanning, provenance).
• Add logging, monitoring and alerting for AI workloads, and write incident runbooks for AI‑specific events.

Secure engineering

• Threat‑model new AI features against the OWASP Top 10 for LLM Applications, MITRE ATLAS and NIST AI RMF.
• Build security checks into CI/CD (SAST, dependency scanning, secret scanning, model scanning).
• Review code and architecture for AI products and vendor AI services.
• Train developers on secure AI patterns and publish reusable libraries and templates.

Key Skills

• LLM & GenAI engineering: LLM APIs (OpenAI, Anthropic, Azure OpenAI, Vertex AI) and open‑weight models; RAG, embeddings and vector databases; agent and tool frameworks such as LangChain or LlamaIndex.
• AI threat mitigation: defending against direct and indirect prompt injection, jailbreaks, system‑prompt and data leakage, data poisoning, model extraction and excessive agency. Mapped to the OWASP Top 10 for LLM Applications and MITRE ATLAS.
• Guardrails & secure by design AI: input and output filtering, PII redaction and policy enforcement around model endpoints. Least‑privilege tool access, sandboxing, human approval for risky agent actions, and tenant isolation, using tools such as NeMo Guardrails, Llama Guard, Lakera or Presidio.
• Application security fundamentals: OWASP Top 10, authentication and authorisation, secrets management, cryptography basics, and secure code review.
• ML fundamentals & supply chain: how models are trained, served and evaluated; enough model understanding to tell model failures from system failures. Securing model weights, datasets and dependencies through signing, scanning (ModelScan) and provenance.
• DevSecOps & CI/CD: building security checks into pipelines, including SAST (Semgrep), dependency and secret scanning (Snyk, Dependabot, Gitleaks), and automated adversarial prompt tests (Garak, PyRIT, Promptfoo).
• Containers & cloud security: Docker and Kubernetes; IAM, networking and secure deployment of AI workloads on AWS, Azure or GCP.

Requirements

• 5‑7 years in software engineering, application security or security engineering, including at least 2 years in a security‑focused role. Must have hands‑on work securing or building at least one LLM or ML application in production.
• Bachelors or Masters degree in Computer Science, Software Engineering, Cybersecurity, Artificial Intelligence or a related discipline, or equivalent practical experience.