Staff+ Researcher, Cybersecurity Products

Capabilities researcher on the team building Claude Security, Anthropic's security product. The researcher works out which security capabilities in frontier models are dependable enough to build on, how they hold up in realistic and adversarial conditions, and where they break, through quick prototypes and careful evaluation. That means designing evaluations based on real security team work, building the datasets, harnesses and scoring behind them, tracking how model security skills change, engaging the security community, and working with engineers to turn promising capabilities into features non-experts can use.

Anthropic wants seven or more years in security research or engineering with deep expertise in a field like offensive security, exploit development, vulnerability research, malware or reverse engineering work, or incident response, plus experience building AI-powered security tools. Based in San Francisco.