Cyber Evaluations Engineer

Cyber Evaluations Engineer in Anthropic's Safeguards organisation who builds and runs the tests that measure how capable its models are at cyber tasks and how well safeguards hold up. The engineer designs capability, uplift and safety evaluations for new models, runs robustness testing before each major release, studies jailbreak and prompt bypass data, designs and tunes probes that detect cyber misuse in production, and works with the cyber policy team on a layered detection architecture, along with the tooling that runs and scores evals.

Anthropic wants experience building evaluations or test suites on tight timelines, hands-on security work such as CTFs, vulnerability research or exploit development, Python, and clear communication of results. Offensive security depth, detection content writing and a Secret clearance or eligibility are pluses. Remote-friendly in the US, with San Francisco and Washington, DC offices.