EPAM SystemsRemote
Lead Security Engineer - AI Red Teaming Engineer
United States of America
Full Time
4 hours ago
H1B Sponsor Likely
Key skills
Adversarial AI testingDirect Prompt InjectionIndirect Prompt InjectionJailbreakingModel behavior manipulationMITRE ATLASMITRE ATT&CKOWASP Top 10Retrieval-Augmented Generation pipelinesAgentic workflows validationPython programmingAI red teaming toolsLLM fundamentalsLLM securityPrompt EngineeringMulti-Agent Systems Evaluation and TestingEnglish proficiencyPythonAIMachine LearningGenerative AILLMRAGAgenticOWASP
About this role
EPAM Systems is seeking a Lead Security Engineer - AI Red Teaming Engineer to focus on the behavioral and algorithmic safety of LLMs and Generative AI systems. The role involves executing prompt injection attacks, identifying model vulnerabilities, and collaborating with cross-functional teams to enhance AI system defenses.
Responsibilities:
- Execute Direct and Indirect Prompt Injection attacks and jailbreaking techniques including multi-turn and roleplay scenarios to test LLM guardrails
- Identify vulnerabilities related to model behavior manipulation and potential data leakage
- Apply MITRE ATLAS and OWASP Top 10 for LLMs frameworks to structure adversarial testing methodologies
- Test Retrieval-Augmented Generation pipelines for data poisoning vulnerabilities
- Validate Agentic workflows against tool-use manipulation, goal hijacking and unauthorized permission escalation
- Develop and script custom boundary-testing payloads using Python
- Leverage AI red teaming and evaluation frameworks to assess model safety and resilience
- Document findings and provide remediation guidance to engineering teams
- Collaborate with cross-functional teams to strengthen defenses across AI systems
Requirements:
- 5+ years of experience in adversarial AI testing including Direct and Indirect Prompt Injection, Jailbreaking and model behavior manipulation
- Familiarity with MITRE ATLAS, MITRE ATT&CK and OWASP Top 10 for LLMs frameworks
- Experience testing RAG pipelines for data poisoning and validating Agentic workflows against tool-use manipulation and goal hijacking
- Proficiency in AI red teaming tools such as PyRIT, GARAK and Giskard or enterprise suites like Confident AI or Repello
- Understanding of LLM fundamentals including tokens, embeddings and context windows
- Skills in Python programming for scripting custom boundary-testing payloads
- Knowledge of LLM security, Prompt Engineering and Security practices
- Background in Multi-Agent Systems Evaluation and Testing
- English proficiency at B2 level or higher
- Expertise in AI Solution Engineering
- Knowledge of Machine Learning