LLM Security Researcher

Crossing Hurdles is seeking a LLM Security Researcher - Red-Teamer to develop adversarial, multi-turn conversations and task-based scenarios for frontier large language models (LLMs). This contract role requires deep expertise in LLM evaluation, rubric design, and adversarial prompting to assess model performance against precise behavioral targets.

Key responsibilities include:

• Designing and refining evaluation rubrics to assess LLM responses with clarity and precision.
• Creating complex, nuanced task packages with transcripts, target behaviors, and rationale.
• Iteratively testing and validating LLM outputs, documenting strengths and failure modes.
• Collaborating with team leads to ensure alignment with evolving project requirements.
• Maintaining meticulous attention to detail and critical thinking to escalate complexity in evaluations.

Qualifications:

• Exceptional written English skills with structured, precise communication.
• Proven experience in AI human data environments (e.g., RLHF, SFT, prompt engineering).
• Deep familiarity with large language models and ability to identify failure patterns.
• Autonomous problem-solving and critical thinking to interpret and execute complex specifications.
• Advantageous: Experience designing evaluation rubrics or adversarial prompting.

This is a remote contract role offering flexible hours (10-40 hrs/week) with compensation ranging from $40 to $65/hour. Ideal for professionals passionate about advancing AI security and model evaluation.