AI security research engineer_
Remote or New York · full-time
Run the experiments that decide how workflows and profiles are built.
What you'll do
Design and run the ablations behind every workflow and profile: which tools, what scope, how many passes, when to verify, which model at which depth. Build the evaluation infrastructure the research team runs on. Read traces, find where the agent went wrong, and fix the harness rather than the finding. Publish the results, including the ones that did not work.
What we look for
You have built agent systems and measured them, not only shipped them. You can read a run log and say what the model was told, what it did, and which of the two was the problem. Python and TypeScript; comfort with evaluation statistics.
What we offer
The research question the company exists to answer — how much of security AI is the harness — as your full-time job, with the budget to run the experiments and the obligation to publish them.