Что будет проверяться
The Frontier Red Team evaluates the cyber and offensive-security capabilities of frontier models, building challenging environments and evaluations to measure what Claude can and cannot do in realistic attack scenarios. The role combines security domain expertise with ML experimentation to inform Anthropic's responsible scaling commitments. A technical interview would probe practical offensive-security knowledge (exploitation, CTF-style problem solving), how to design rigorous and non-gameable capability evaluations, and building RL or agentic environments that elicit and measure a model's security-relevant behavior.
ExoForm не аффилирован с Anthropic. Это независимая тренировочная страница.