The posting
The AI Data and Safety team plays a critical role in advancing Seed foundational models across modalities and improving AI-native applications built on the Seed model series. We work across the data lifecycle, from defining evaluation approaches and translating user feedback and benchmark outcomes into data requests, to building scalable processes that improve data quality and support rapid model iteration.
Our team combines technical and operational capabilities, bringing together multidisciplinary talent across product management, data engineering, and data operations. Our work is driven by people who think deeply about model behavior, move quickly to solve complex problems, and bring first-hand experience as both builders and users of models and agents.
In close partnership with Seed researchers, industry experts, and leading data vendors, we tackle challenging data problems at the frontier of AI development, helping improve both model performance and user experience.
Your job is to turn ambiguous concerns into clear, testable problems: - Work with researchers and product managers to research emerging AI safety risks using industry research, academic work, real-world cases, and expert input - Define risk mechanisms, scope, boundaries, and representative failure modes - Translate risk definitions into evaluation cases, benchmarks, metrics, and thresholds - Run exploratory analyses, pilots, and lightweight experiments - Review raw cases and evaluation results, identify failure patterns, and iterate the methodology
Prior AI Safety experience is not required. We are open to candidates from research, behavioral science, experimentation, risk, quality, or other adjacent backgrounds who have demonstrated strong problem-solving and measurement skills.
Minimum Qualifications - Minimum 2 years of professional or research experience in an analytical, research, strategy, evaluation, or similarly problem-solving-oriented role. - Strong examples of end-to-end / 0-1 ownership of an ambiguous problem: you independently clarified what the problem was, gathered the evidence needed to understand it, and turned it into a concrete output others could use—such as a taxonomy, metric, experiment, test framework, evaluation set, decision rule, or research recommendation. - Hands-on research or analysis experience using raw data, cases, user research, literature, experiments, or other primary evidence. - Strong written and verbal reasoning: able to explain what you concluded, what evidence supports it, what remains uncertain, and why you chose one approach over alternatives.
Preferred Qualifications We welcome candidates with different methodological strengths. You may have one or more of the following: - Social/behavioral research strengths: experience in behavioral science, HCI, UX research, consumer research, policy research, integrity research, or similar fields; Strong in construct definition, human behavior, qualitative/quantitative research, or experiment design. - Technical/systems evaluation strengths: experience in search/recommendation quality, experimentation, fraud/risk strategy, QA/test strategy, data quality, model evaluation, reliability, or similar fields; Strong in test design, failure analysis, metrics, and root-cause analysis. - Familiarity with GenAI models, AI evaluation, or AI safety. - SQL, Python, Agents, or other lightweight technical skills that support independent investigation. - Evidence of successfully learning or moving across domains.
