Associate, Digital Operations
Quick Summary
· Act as a specialist individual contributor executing AI red Teaming adversarial evaluation pilot. · Formulate risk hypotheses, design creative adversarial scenarios, and execute multi-turn,
Time Type:Full time2026-12-19
Job Title:
Associate, Digital OperationsJob Description
ROLE: LLM / AI Red-Team EvaluatorDESIGNATION: LLM / AI Red -Team Evaluator
YEARS OF EXPERIENCE: 3 plus years
Concentrix is a technology-enabled global business services company specializing in customer engagement and improving business performance. We partner with ambitious, progressive executives around the world to future-proof their business and stay ahead of the competition and customer expectations
Responsibilities
~1 min read· Act as a specialist individual contributor executing AI red Teaming adversarial evaluation pilot.
· Formulate risk hypotheses, design creative adversarial scenarios, and execute multi-turn, adaptive prompt attacks on authorized systems.
· Critically evaluate model outputs for grounding errors, source contradictions, stale information, and trust-boundary violations (such as indirect prompt injections).
· Capture, package, and document complete, reproducible evidence including exact prompt sequences, session/version context, and timestamps.
· Submit detailed finding cards for QA and calibrate judgments based on updated rubrics and guidelines.
· Separately classify candidate findings, quality issues, and inconclusive outcomes, knowing exactly when to stop and escalate unsafe or uncertain cases.
· Meet or exceed pre-production certification targets (including achieving a ≥85% agreement rate on adjudicated practice sets with zero material omissions).
· Maintain strong ownership of assigned attack surfaces and test surfaces within client-approved boundaries.
· Stay informed of client taxonomies, system changes, policy revisions, and evolving risk definitions.
· Collaborate with cross-functional teams including the Red-Team Team Lead, Project Manager, and QA analysts to resolve challenges constructively.
· Core Technical Competencies: AI Red Teaming, Adversarial AI, Prompt Injection, Jailbreak Testing, Responsible AI, AI Governance, AI Safety.
· Core Methodology Competencies: LLM Evaluation, Model Evaluation, RLHF (Reinforcement Learning from Human Feedback), Human Feedback, Prompt Engineering, GenAI Testing, AI QA.
· Strong analytical, skeptical, and troubleshooting capability across LLM/RAG safety environments.
· Excellent written and verbal communication skills in US-English for structured reporting.· High attention to detail and structured problem-solving approach to isolate false positives and verify citations.
· Demonstrated ability to build original, bounded attack hypotheses and execute multi-turn context escalation scenarios.
· Quality-focused mindset with accountability for complete evidence and defensible coverage over pure prompt volume.
· Defensible Coverage and Finding Reproducibility
· QA Calibration and Adjudication Scores (≥85% Agreement Target)
· Client Boundary & Privacy Policy Compliance
· Execution of Structured Finding Cards and Calibration Updates
'Neither Concentrix nor any authorized 3rd party who assist with our recruitment process, ever ask candidates for ‘recruitment’, ‘processing’ or any other kind of fees in exchange for offer letters from Concentrix. Offer letters and other recruiting correspondence from Concentrix are printed on Concentrix letter head with authentic signatures of appropriate Concentrix authorities.'
Location:
IND Hyderabad - Unit No. 201, 2nd Flr, Building 4 at MindspaceLanguage Requirements:
Time Type:
Full time2026-12-19Location & Eligibility
About this job
- Posted
- Today
Listing Details
- Posted
- November 20, 2026
- First seen
- November 20, 2026
- Last seen
- November 20, 2026
Posting Health
- Days active
- 0
- Repost count
- 0
- Trust Level
- 55%
- Scored at
- November 20, 2026
Signal breakdown
Stay ahead of the market
Get the latest job openings, salary trends, and hiring insights delivered to your inbox every week.
No spam. Unsubscribe at any time.