Data Assistant Contract Role: Help Advance AI Safety Through Precision Labeling

At the forefront of applied AI safety, our team works to ensure that advanced AI systems protect and empower people—not harm them. We believe that safeguarding mental health and human agency is just as essential as preventing technical failure. If you’re driven by real-world impact and have a keen eye for detail, this short-term role offers a chance to contribute to the foundational work that enables large-scale behavioral risk detection in AI systems.

We are seeking a short-term contractor (3 months) to support an AI Lab safety team in preparing datasets for future automation and analysis. This role centers on data labeling and organization, with a focus on flagging behavioral patterns in historical AI data. Your work will directly support efforts to improve detection of safety-relevant behaviors—such as self-harm, manipulation, and emotional escalation—in large language models.

This role is ideal for someone who enjoys methodical, detail-oriented work, is comfortable navigating large datasets, and can work within a technical environment. Basic SQL and experience using Jupyter notebooks are required, and Python proficiency is a plus.

Please note: This work may involve reviewing text that references sensitive topics, including mental health struggles, suicidal ideation, addiction, or interpersonal violence. Candidates should be comfortable reading and labeling this kind of material.

• Label historical conversation data: Use internal tools and clear guidelines to tag safety-relevant behaviors in AI data, including signs of harmful behavior.
• Query and organize datasets: Use basic SQL queries to extract and review relevant data for labeling and analysis.
• Work in Jupyter notebooks: Navigate and run your work within Jupyter notebooks to support traceability and collaboration with technical teams.
• Collaborate with cross-functional partners: Work alongside analysts, researchers, and engineers to refine labeling criteria and surface edge cases.
• Support labeling tool improvements: Provide structured feedback on labeling workflows to help improve future annotation tools and processes.

• Have a meticulous eye for detail and enjoy repetitive but high-stakes tasks where accuracy matters.
• Are organized and systematic, with a track record of following annotation guidelines and managing large volumes of information consistently.
• Are comfortable writing and interpreting basic SQL queries to retrieve and filter relevant data.
• Have hands-on experience using Jupyter notebooks to review data and document workflows.
• Are curious and technically inclined, with some exposure to Python or a willingness to learn.
• Communicate clearly and take initiative when something looks off or doesn’t align with the guidelines.
• Care about mental health, safety, and ethical tech—and want to support systems that take these seriously.

This is a great opportunity to support cutting-edge safety work in AI while building experience with real-world data annotation workflows in a technical research setting.