Это чётко описанная роль на стыке значимых исследований безопасности ИИ и production engineering в сильной AI-компании; при этом оценить оплату по доступному рыночному бенчмарку нельзя.
Передовой ИИ
Чёткая зона ответственности
Значимая миссия
?Возможны дежурства
Мэтч & Сопровод
Для мэтча с этой вакансией нужен Plus
Подключить Plus
Описание вакансии
Текст:
Hirify AI/Оригинал
TL;DR
Research Engineer, AI Monitoring (AI/Safety Classifiers): Building and operating real-time classifiers that detect unsafe actions by AI agents across research, training, evaluations, and internal deployments with an accent on high-recall and high-precision detection, reliability, latency, and cost. Focus on designing adversarially robust monitoring systems, building datasets and evaluation pipelines, and integrating low-latency classifiers into critical production paths.
Location: London, UK; hybrid policy requiring all staff to work from one of the offices at least 25% of the time, with some roles requiring more office time
Annual salary: £1–£1 GBP
Company
Anthropic is a public benefit corporation developing reliable, interpretable, and steerable AI systems, with a focus on safe and beneficial AI.
What you will do
Design, train, and evaluate classifiers, prompted models, trained classifiers, and model-internal probes that detect unsafe agent actions in real time.
Deploy, tune, and operate classifiers across internal agents, reinforcement learning training, and evaluations while owning accuracy, reliability, latency, cost, and failure behavior.
Build datasets, labeling pipelines, evaluations, and harm taxonomies to measure and improve monitoring systems.
Partner with red-teaming colleagues to stress-test classifiers and harden them against adversarial findings.
Work with agent, environment, and evaluation teams to resolve false positives without weakening protection.
Contribute to technical direction, partnerships, and investment priorities.
Requirements
Ability to organize and prioritize several parallel workstreams and communicate clearly with affected teams.
Ability to critically review code and experiment results, including work produced by AI agents.
Sincere interest in developing safe and beneficial AI systems.
Bachelor’s degree or equivalent combination of education, training, and experience.
Relevant field of study demonstrated through coursework, training, or professional experience.
Required years of experience correlate with the internal job-level requirements.
Nice to have
Experience leading ambiguous technical projects and setting direction for engineers or researchers.
Experience evaluating large language models, developing classifiers, or operating high-scale production systems in changing environments.
Experience building detection systems with rare true positives and costly false positives, such as abuse, fraud, spam, or intrusion detection.
Experience running low-latency, high-availability services, including on-call responsibility.
Experience fine-tuning language models or using probes and other methods to read model internals.
Experience with LLM inference or serving infrastructure, or integrating components into training infrastructure.
Security background, including sandboxing or detection and response.
Familiarity with AI control, scalable oversight, agent safety, or reinforcement learning research.
Culture & Benefits
Collaborative research environment focused on a small number of large-scale AI research efforts.
Frequent research discussions and emphasis on communication, impact, and trustworthy AI.
Competitive compensation and benefits, optional equity donation matching, generous vacation and parental leave, and flexible working hours.
Office-based collaboration with staff expected to attend an office at least 25% of the time.
Visa sponsorship may be available; sponsorship is assessed by role and candidate, with immigration-lawyer support.
Candidate AI usage is governed by Anthropic’s application-process 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 →