Назад
10 дней назад

Staff+ Software Engineer, Safeguards Evals (AI)

320 000 - 485 000$
Формат работы
hybrid
Тип работы
fulltime
Грейд
senior
Английский
b2
Страна
US
Вакансия из списка Hirify.GlobalВакансия из Hirify Global, списка международных tech-компаний
Для мэтча и отклика нужен Plus

Мэтч & Сопровод

Для мэтча с этой вакансией нужен Plus

Описание вакансии

Текст:
/
TL;DR
Staff+ Software Engineer, Safeguards Evals (AI): Building evaluation methods, datasets, harnesses, and production pipelines for LLM behavior and agentic safety investigations with an accent on representative abuse data, grading accuracy, and robust measurement. Focus on designing long-horizon agent evaluations, constructing reinforcement-learning environments, analyzing complex harm areas, and productionizing safeguards for training and launch decisions.

Location: San Francisco, CA or Seattle, WA. Hybrid policy: staff are expected to work from one of the offices at least 25% of the time.

Annual salary: $320,000–$485,000 USD.

Company

Anthropic is a public benefit corporation building reliable, interpretable, and steerable AI systems intended to be safe and beneficial for users and society.

What you will do

  • Design and run experiments that improve evaluation quality, including representative test-data generation, realistic user-behavior simulation, grading validation, and coverage analysis.
  • Build and own evaluation harnesses for agentic investigation systems, defining metrics, test cases, grading approaches, and end-to-end performance measurements.
  • Construct reinforcement-learning environments and drive improvements in difficult AI safety and harm areas.
  • Build evaluation datasets grounded in real-world misuse, including cyber attacks, biological weapons, and influence operations, using real traffic patterns and synthetic generation.
  • Collaborate with Policy and Enforcement to convert observed harm patterns into measurable evaluations.
  • Productionize successful research in evaluation, regression, and release pipelines and build tools that let policy experts create and iterate on evaluations without engineering support.

Requirements

  • Experience building and maintaining data pipelines.
  • Experience working with LLMs and understanding their capabilities and failure modes, especially agentic systems using tools and multi-step reasoning.
  • Strong data analysis skills with the ability to draw reliable insights from large datasets.
  • Ability to move between research prototypes and production-quality code.
  • Ability to turn ambiguous problems into concrete, testable experiments.
  • Strong commitment to AI safety and meaningful impact.
  • Bachelor’s degree or an equivalent combination of education, training, and experience in a relevant field demonstrated through coursework, training, or professional experience.

Nice to have

  • 8+ years of industry software engineering or ML engineering experience.
  • Experience building or contributing to LLM or agent evaluation frameworks, benchmarks, or automated grading systems.
  • Experience in trust and safety, content moderation, or abuse detection systems.
  • Experience with red teaming, adversarial testing, or AI-system jailbreak research.
  • Experience with synthetic data generation or data augmentation.
  • Experience with distributed systems or large-scale data processing.
  • Experience with prompt engineering or LLM-powered applications.

Culture & Benefits

  • Collaborative research environment focused on large-scale efforts in trustworthy and steerable AI.
  • Frequent research discussions and strong emphasis on communication.
  • Flexible working hours, generous vacation and parental leave, competitive benefits, and optional equity donation matching.
  • Visa sponsorship is available, although sponsorship cannot be guaranteed for every role or candidate.
  • Office-based collaboration is required at least 25% of the time under the hybrid policy.

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 →