19 Nov
|
Wipro
|
Bengaluru
AI Penetration Testing Engineer
Cybersecurity and Risk Services
Cyber Defense | Threat and Vulnerability Management | Offensive Security and AI / LLM
Security track
Document control Detail
Document ID CRS-TVM-JD-AIPT
Version v1.0
Date 12 October 2026
Owner TVM Practice (Practice Delivery Lead)
Review date 12 October 2026
Approval status Approved
Classification Internal - General Use
Cybersecurity and Risk Services | AI Penetration Testing Engineer -
Internal - General Use CRS-TVM-JD-AIPT | v1 Page 2 of 3
- Role snapshot
Attribute Detail
Role AI Penetration Testing Engineer
Practice / cluster TP - Cyber Defense - TVM | Offensive Security skill cluster with AI / LLM
Security specialization
Experience range 5+ years overall in offensive security / penetration testing
Hands-on web, API and cloud penetration testing experience (mandatory)
Demonstrated hands-on security testing of AI / LLM-based applications
(mandatory)
AI transformation zone Zone 3 (current role created through AI infusion)
Role family Offensive Security and Penetration Testing track | AI / LLM Security
(emerging tech)
Location / engagement model To be confirmed per demand (offshore / onsite / hybrid)
- Role purpose
- Perform authorized, scoped penetration testing and adversarial assessment of AI systems,
including LLM applications, AI agents, retrieval-augmented generation (RAG) pipelines, and the infrastructure that supports them.
- Identify, validate, and demonstrate the business impact of AI-specific vulnerabilities alongside conventional application, API, and cloud weaknesses.
- Translate findings into practical, prioritized remediation and guardrail recommendations for engineering and data science teams.
- Build repeatable AI security testing methodology, test-case libraries, and automated adversarial evals that scale across engagements.
- Use governed AI tooling to accelerate offensive workflows, with human validation of every finding.
- Key responsibilities
3.1 Engagement planning and threat modeling
- Scoping: Scope AI security engagements, confirm authorization, and agree rules of engagement, test windows, environments, and safety constraints.
- Threat modeling: Threat model AI systems, covering trust boundaries, data flows, model and tool interactions, and abuse cases, using frameworks such as MITRE ATLAS.
- Test planning: Build risk-based test plans mapped to the OWASP Top 10 for LLM Applications and the system's architecture.
3.2 LLM and GenAI application testing
- Injection and manipulation: Test for direct and indirect prompt injection, jailbreak resistance,
and system-prompt leakage.
- Data exposure: Assess sensitive information disclosure, training-data and context leakage,
and cross-user or cross-tenant data exposure.
Cybersecurity and Risk Services | AI Penetration Testing Engineer -
Internal - General Use CRS-TVM-JD-AIPT | v1 Page 3 of 3
- Output handling: Evaluate improper output handling where model output reaches downstream components (e.g., rendering, code execution, queries).
- Guardrail testing: Test content-safety and policy guardrails for bypass and inconsistent enforcement.
- Availability and cost abuse: Assess resource-exhaustion and cost-abuse risks (unbounded consumption).
3.3 Agents, RAG and integrations
- Test agentic systems for excessive agency, unsafe tool use, privilege escalation through tools and plugins, and insufficient human approval steps.
- Assess RAG pipelines for data and embedding poisoning, retrieval manipulation, and access-
control gaps in vector stores and knowledge sources.
- Test plugin, connector, and Model Context Protocol (MCP)-style integrations for authentication,
authorization, and input-validation weaknesses.
3.4 Model and ML supply chain
- Review model provenance, third-party model and dataset sourcing, and model registry and artifact integrity.
- Assess risks of model tampering, unsafe model serialization formats, and dependency vulnerabilities in ML frameworks.
- Evaluate adversarial ML risks (evasion, extraction, membership inference) where in scope and authorized.
3.5 Supporting application, API and cloud testing
- Test the web, API, identity, and cloud layers hosting AI workloads, including authentication,
authorization, secrets, and storage configuration.
- Assess AI platform services (e.g., managed model endpoints and AI studios) for misconfiguration and excessive permissions.
- Chain AI-specific and conventional weaknesses into realistic attack paths that demonstrate business impact safely and reproducibly.
3.6 Reporting, remediation and retesting
- Produce reports with executive summaries, technical detail, reproduction steps, evidence, and risk ratings.
- Recommend layered mitigations: input / output filtering, privilege separation, human-in-the-
loop controls, retrieval access controls, and monitoring.
- Lead technical readouts with engineering and data science teams; validate fixes through retesting and regression test cases.
3.7 Automation, research and enablement
- Build and maintain adversarial test-case libraries and automated eval suites, and integrate them into CI/CD for pre-release AI security testing.
- Track emerging AI attack techniques, frameworks, and regulations; turn research into methodology updates.
- Support detection engineering by sharing AI misuse patterns with SOC teams for monitoring use cases.
- Mentor engineers on AI security testing; contribute playbooks and reusable accelerators.
- Support pre-sales and solutioning with AI security testing approach and effort inputs where required.
Cybersecurity and Risk Services | AI Penetration Testing Engineer -
Internal - General Use CRS-TVM-JD-AIPT | v1 Page 4 of 3
- AI-infused responsibilities (Zone 3)
- Use AI to generate test checklists and adversarial test cases from scope, architecture, and threat models; validate every case manually.
- Use AI-assisted tooling to scale prompt-based testing and cluster results; confirm findings through hands-on reproduction.
- Use AI to map findings to attack paths and business impact narratives; validate assumptions with evidence.
- Use AI to draft report sections and remediation options with trade-offs; ensure accuracy and reproducibility.
- Maintain internal knowledge bases of approved prompts, test cases, and reporting templates under access control.
Guardrails
- Never copy AI-generated payloads or scripts without understanding and peer review; AI must not replace creative adversarial thinking.
- Use only enterprise-approved / private AI endpoints; no client data, prompts, model outputs,
credentials, or internal URLs in unapproved tools.
- Keep all testing authorized, scoped, and non-destructive; stop and escalate if a test risks unintended impact.
- Key deliverables
5.1 Planning
- Signed-off scope, rules of engagement, and test-environment confirmation per engagement.
- AI system threat model with trust boundaries, data flows, and abuse cases.
- Risk-based AI security test plan mapped to OWASP LLM Top 10 and MITRE ATLAS.
5.2 Assessment
- Assessment report: executive summary, technical findings, evidence, reproduction steps, and risk ratings.
- Attack-path narratives showing how AI-specific and conventional weaknesses chain to business impact.
- Prioritized remediation and guardrail recommendations with short-term and long-term options.
- Retest report and closure verification notes.
5.3 Reusable assets
- Adversarial test-case library for LLM apps, agents, and RAG systems.
- Automated AI security eval suites and CI/CD integration guidance.
- AI security testing methodology, playbooks, and reporting templates.
- Detection use-case inputs for SOC monitoring of AI misuse.
5.4 Enablement
- Technical readouts and knowledge-transfer sessions for engineering and data science teams.
- Training content and mentoring plans for AI security testing.
Cybersecurity and Risk Services | AI Penetration Testing Engineer -
Internal - General Use CRS-TVM-JD-AIPT | v1 Page 5 of 3
- Experience and qualifications
6.1 Mandatory
- 5+ years of experience in offensive security / penetration testing, with strong web, API, and cloud testing depth.
- Demonstrated hands-on security testing of AI / LLM-based applications (e.g., prompt injection,
data leakage, agent and tool abuse).
- Working knowledge of how LLM applications are built: prompts, context windows, embeddings,
RAG, agents, and tool calling.
- Strong scripting ability in Python for test automation and tooling.
- Bachelor's degree in Computer Science, IT, Cybersecurity, or a related field (or equivalent demonstrated experience).
6.2 Desirable
- Experience with adversarial ML techniques and ML frameworks (e.g., PyTorch, TensorFlow,
Hugging Face).
- Experience with AI platform services on at least one cloud (Azure AI, AWS Bedrock /
SageMaker, Google Vertex AI).
- Red-team, purple-team, or detection-aware testing experience.
- Published research, CVEs, conference talks, or responsible-disclosure contributions in AI security.
- Customer-facing delivery, readout, and solutioning experience.
- Technical skills and tools
Area Skills / typical tools
LLM red teaming e.g., PyRIT, garak, promptfoo, Promptmap; custom prompt-injection and jailbreak test harnesses
Adversarial ML e.g., Adversarial Robustness Toolbox (ART), TextAttack, Counterfit
Web and API testing Burp Suite Pro and extensions, OWASP ZAP, Postman; authenticated and
API-mode testing
Cloud and AI platforms Azure AI / Azure OpenAI, AWS Bedrock / SageMaker, Google Vertex AI;
cloud IAM and configuration review
AI application stacks LangChain, LlamaIndex, Semantic Kernel, agent frameworks, vector databases
Model ecosystem Hugging Face, model registries, model serialization formats and scanning
(e.g., ModelScan)
AI security posture AI-SPM and AI firewall / guardrail platforms (e.g., Wiz, Prisma AIRS, Azure AI
Content Safety)
Automation Python, Jupyter, REST APIs; CI/CD integration of eval suites
Identity and infrastructure OAuth / OIDC, JWT, secrets management; container and Kubernetes basics
Reporting Evidence management, report templates, knowledge bases
AI enablement Approved AI copilots, structured prompting, AI-output validation
Cybersecurity and Risk Services | AI Penetration Testing Engineer -
Internal - General Use CRS-TVM-JD-AIPT | v1 Page 6 of 3
- Knowledge areas and standards
- OWASP Top 10 for LLM Applications and OWASP GenAI Security Project guidance.
- MITRE ATLAS (Adversarial Threat Landscape for AI Systems) and MITRE ATT&CK.
- NIST AI Risk Management Framework (AI RMF) and NIST AI 100-2 (adversarial machine learning taxonomy).
- OWASP Top 10 (Web, API), OWASP WSTG, and penetration testing methodologies (PTES,
NIST SP 800-115).
- AI regulation and standards awareness: EU AI Act and ISO/IEC 42001.
- CVSS, CWE, and severity rating approaches adapted to AI-specific findings.
- Responsible AI principles and secure AI development lifecycle practices.
- Success measures
- Coverage of in-scope AI systems tested before and after release.
- High-quality, reproducible findings with low false positives and clear business impact.
- Timely remediation and verified closure of critical and high AI findings.
- Reduced repeat findings through effective guardrail and design-level recommendations.
- Growth and reuse of AI test-case libraries and automated eval suites.
- Zero testing activity outside approved scope or causing unapproved impact.
- Positive stakeholder feedback on reports and readouts.
📌 AI Penetration Testing Engineer (Bengaluru)
🏢 Wipro
📍 Bengaluru