Llm security evaluation expert

Hello
Greetings from Zetta Mine Labs Pvt Ltd!!
We are looking for LLM Security Evaluation Expert for PAN India.
Job Role : LLM Security Evaluation Expert
Location : PAN India
Notice Period : NA( As it is for Gig Workers)
Experience : 3 to 6 years
Relevant Exp : Minimum of 3 Years Experience in below list Skills
✓ LLM architecture & training process knowledge
✓ OWASP LLM Top 10 & MITRE ATLAS
✓ Adversarial prompt design & execution
✓ Vulnerability assessment for AI/ML systems
✓ Prompt injection & jailbreaking techniques
✓ Offensive security tooling & methodology
✓ Role-playing and bias exploitation attacks
✓ Security reporting and findings documentation
Mandatory :
3+ years of experience in offensive security, AI/ML security research, penetration testing of AI systems, or dedicated LLM red-teaming.
Deep working knowledge of LLM architecture, training processes, and alignment techniques (RLHF, fine-tuning, instruction-following).
Proven hands-on expertise in adversarial prompt design, jailbreaking, prompt injection, and role-play attack methodologies.
Strong grounding in cybersecurity principles applied to AI/ML systems; familiarity with OWASP LLM Top 10 and related frameworks.
Clear and precise written communication skills — ability to document technical findings thoroughly and present security risks to non-technical stakeholders.
Commitment to responsible testing practices and understanding of the ethical implications of AI security research.
Job Description:
As a LLM Security Evaluation Expert, you will be the adversarial mind that frontier AI systems need but rarely have. Your mandate is to rigorously probe large language models for security vulnerabilities — designing, executing, and documenting adversarial prompt campaigns that expose real failure modes before bad actors find them first. You will operate at the intersection of offensive security and AI research , crafting sophisticated attack vectors including prompt injection, jailbreaking, role-play manipulation, and bias exploitation, and translating findings into actionable security intelligence for AI development teams.
This is not a compliance checkbox role. It requires genuine adversarial creativity, deep LLM knowledge, and the discipline to document what you find with rigour and ethical clarity.
Key Responsibilities:
Adversarial Prompt Engineering & Attack Design
Develop and execute a structured suite of adversarial prompts — from foundational to highly sophisticated — targeting known and emerging LLM vulnerability classes
Design attacks to bypass safety filters and content moderation policies; induce leakage of sensitive or confidential information; manipulate outputs toward harmful, biased, or unintended content
Engineer prompt injection, jailbreaking, role-playing, and model bias exploitation scenarios; test against a spectrum of LLM architectures and deployment configurations
Security Evaluation & Vulnerability Analysis
Systematically execute adversarial test campaigns against target LLMs; analyse responses to identify successful exploits, failure patterns, and exploitable security weaknesses
Conduct vulnerability assessments against published frameworks including OWASP LLM Top 10, MITRE ATLAS, and NIST AI RMF; document findings with reproducible methodology
Assess the resilience of safety mechanisms including RLHF alignment, Constitutional AI, and fine-tuning guardrails against adversarial pressure.
Reporting, Standards & Collaboration
Produce clear, technically precise vulnerability reports including exploit documentation, impact assessment, and remediation recommendations
Collaborate with AI development and red-team leads to translate findings into security requirements, test standards, and policy updates
Contribute to Platina’s growing adversarial evaluation library and internal red-teaming playbooks.
Please provide following detail along with your updated profile to
Primary Skill -
Total Experience -
Relevant Experience :
Current location
Highest Qualificaiton:
Expertise/Stream:
Domain :
I will be reaching out to you over a call or email as soon as possible.
Thanks & Regards,
Vyshnavi