Research Engineer — AI Alignment & Evaluation

W3 Global Sourcing•San Francisco, CA

5d•Hybrid

About The Position

We are representing a high-growth AI research organization working at the intersection of frontier model evaluation, AI safety, and security. The team develops sophisticated evaluation environments designed to surface undesirable or misaligned model behavior and help leading AI organizations better understand how advanced systems behave under complex, long-horizon conditions. This is a technically rigorous environment for engineers who are interested in AI alignment, agent behavior, model evaluation, and building systems that help make increasingly capable AI more reliable and controllable.

Requirements

1+ years of experience in software engineering, machine learning engineering, technical research, or a closely related field.
Strong traditional software engineering fundamentals.
Proficiency with Python.
Strong interest in AI alignment, AI safety, or AI security.
Ability to reason carefully about complex systems and ambiguous failure modes.
Strong conceptual judgment and the ability to think through how an autonomous agent may interpret or exploit a task.
Ability to learn new technical domains quickly.
Experience using LLMs or AI agents effectively as part of technical workflows.
Strong ability to assess whether agent-generated work is correct, including when errors are subtle.
Comfortable taking full ownership of technically demanding projects with limited oversight.
High standards for quality, execution, and accountability.

Nice To Haves

Experience building evaluation frameworks, benchmarks, simulation environments, or agent-based systems.
Exposure to frontier language models or autonomous agent workflows.
Background in AI safety, alignment research, adversarial testing, or security.
Experience designing tasks that require multi-step or long-horizon reasoning.
Research experience involving model behavior, reward hacking, robustness, or control mechanisms.

Responsibilities

Design and build complex evaluation environments for frontier AI models.
Own evaluation projects end to end, including ideation, implementation, testing, grading, measurement, and iteration.
Investigate potential model failure modes and identify ways advanced agents may exploit or circumvent intended constraints.
Develop and improve software infrastructure used to isolate, reproduce, and evaluate model behavior.
Work extensively with LLM-based agents to accelerate implementation and research workflows.
Review agent-generated work critically and identify subtle technical or conceptual errors.
Build long-horizon tasks that operate near the edge of current model capabilities.
Apply strong qualitative judgment when evaluating behavior that cannot be captured through simple automated metrics.
Rapidly learn unfamiliar technical domains as required by individual evaluation environments.
Share findings, lessons, and technical context with a highly collaborative research and engineering team.