Model Safety Investigations Specialist (AI)
Мэтч & Сопровод
Для мэтча с этой вакансией нужен Plus
Описание вакансии
Location: Remote, United States-based; primary working hours are Pacific Time. Candidates in Washington, DC, New York City, the San Francisco Bay Area, or Denver are expected to work from a office at least two days per week. In-person onboarding is required in Washington, DC, New York City, or San Francisco during the first week; occasional business travel may be required.
Salary: $120,000–$150,000 annually, depending on experience and location.
Company
provides safety and threat-intelligence capabilities, including adversarial red teaming, model evaluations, and intelligence collection, for AI labs, technology companies, and enterprise platforms.
What you will do
- Triage inbound model safety reports, check for duplicates and relevant history, and gather minimum evidence with approved tools.
- Use logs and case evidence to investigate unexpected model behavior and document observed facts and remaining uncertainty.
- Apply severity and labeling guidance consistently, distinguishing observed behavior and reported harm from hypotheses about cause, intent, or prevalence.
- Write concise case summaries covering evidence, uncertainty, rationale, and recommended next steps; promptly escalate time-sensitive or high-severity cases.
- Label cases, participate in calibration and quality reviews, and identify recurring patterns, ambiguous guidance, and detection gaps.
- Track case complexity, handling time, and escalations to support coverage and staffing improvements.
Requirements
- 2–5 years of hands-on experience investigating sensitive technical cases in trust and safety, incident response, fraud, abuse, or a similar operational setting.
- Ability to reconstruct scenarios from incomplete or conflicting evidence and write concise, defensible summaries that another reviewer can verify.
- Ability to apply policy or severity guidance consistently, recognize cases exceeding personal authority, and escalate without overstating evidence.
- Ability to manage a queue without sacrificing accuracy, incorporate calibration feedback, and follow strict privacy and access controls.
- Comfort interpreting technical information such as logs and piecing together evidence from multiple sources.
- Ability to handle sensitive or distressing content with discretion, sound judgment, and careful handling of user information.
Nice to have
- Experience reviewing AI conversations or investigating model behavior, including jailbreaks, unsafe outputs, or unexpected agent actions.
- Experience with case labeling, reviewer calibration, or quality assurance.
- Familiarity with basic SQL or approved AI-assisted analysis tools.
Culture & Benefits
- 9-month position with potential for extension.
- Fully remote work model for candidates based in the United States, with Pacific Time working hours.
- Comprehensive health, dental, and vision coverage.
- Generous paid time off and paid holidays.
- 401(k) plan.
Будьте осторожны
Если работодатель просит войти в их систему, используя iCloud/Google, прислать код/пароль, запустить код/ПО, не делайте этого. Это мошенники. Обязательно жмите «Пожаловаться» или пишите в поддержку. Подробнее в гайде →