Generative AI & RLHF

Generative AI & RLHF

This worldwide contractor role requires 20+ hours per week and strong application security experience.

The work

You will create and evaluate security tasks for generative AI applications, including agents and retrieval-augmented generation systems. The work covers vulnerable and patched implementations, authorization design, code-level findings, and tests that show whether security fixes work.

You will source or build GenAI applications, create vulnerable versions and patched counterparts, and trace data and action flows across files, callbacks, registries, and background jobs. You will also use a live model to demonstrate relevant paths when practical.

• Define authorization contexts for actors, trust boundaries, tenants, permitted data and actions, purposes, recipients, and document-level access.
• Document findings with exact code spans, traps, control status, and remediation recommendations.
• Write exploit and reproduction tests that demonstrate vulnerable behavior and confirm that patches prevent it.
• Evaluate issues such as sensitive information disclosure, excessive agency, prompt injection, tool abuse, and logic or authorization flaws that scanners may miss.

What it pays and takes

This is a part-time contractor role open worldwide. The listing does not provide a pay rate.

• Time requirement: 20+ hours per week.
• Location: Open worldwide.
• Language: Clear technical English is required.
• Required experience: At least three years in application or product security, secure code review, penetration testing, or vulnerability research.
• Required knowledge: Data-flow reasoning, authentication, authorization, least privilege, tenant isolation, trust boundaries, and approval flows.
• Required technical background: Production coding experience and knowledge of the OWASP Top 10 for LLM Applications.
• Valuable experience: CWE mapping, exploit and reproduction tests, CodeQL or Semgrep, SARIF, LLM tool calling, agents, MCP, RAG, memory, tracing, LangChain, LlamaIndex, relevant SDKs, OpenTelemetry, Docker, and Git.
• Helpful background: SWE, OSCP, GWAPT, or CSSLP credentials; CVE credits; bug bounty work; published security research; AI red-teaming; or LLM evaluation experience.

How it works

You apply on OpenTrain with your resume, then complete the application on the hiring site.

About AI training work

AI training is the human work behind systems that learn from examples, reviews, tests, and feedback. Security evaluators help identify unsafe behavior and improve how generative AI systems handle data, tools, permissions, and user requests.

Requirements

• Languages: English