SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Salary: USD 320,000 - 405,000 / annual
Anthropic is seeking a Capabilities Researcher to join the Claude Security team. In this role, you will identify which security capabilities in frontier AI models are ready for production, measure their performance rigorously, and work to make them accessible to non-expert security teams.
You'll be responsible for fast prototyping to define the AI frontier for cybersecurity work, designing evaluations that measure model performance on real-world security tasks, and building the datasets and scoring harnesses those evaluations require. You'll engage with the cybersecurity community to understand where AI can have the most impact, then partner with engineers and researchers to operationalize promising capabilities into reliable customer-facing products.
Key responsibilities include: prototyping rapidly to explore AI capabilities for security; designing rigorous evaluations that measure performance on actual security team workflows; building datasets and evaluation infrastructure; engaging with the cybersecurity community; working with engineers to turn research into product; tracking how model capabilities evolve; and communicating findings to inform product strategy.
You should have deep expertise in at least one security domain—vulnerability research, exploit development, reverse engineering, malware analysis, incident response, or offensive security. You're comfortable moving quickly from idea to prototype, designing honest evaluations, and communicating technical findings clearly. 7+ years of experience in security research or engineering is required.
Strong candidates may also have published research, CTF results, CVEs, or open-source security tools; experience with model evaluation, benchmarking, or red teaming; experience building agentic applications; or familiarity with AI safety considerations in security contexts.
This is a research role on a product team with broad latitude to investigate what interests you. It suits someone with concrete ideas about what AI should do for security teams and who wants the models, time, and engineering support to pursue them.
