The company builds the environments, tasks and evaluations that AI teams use to train and test their agents.
Its benchmark includes 100 audit-and-patch tasks across 54 weakness types, in which agents must find and fix vulnerabilities in real codebases. Artificial Analysis includes it in its Cyber Index.
The company works with frontier labs on their models' cyber capabilities and is now bringing the same defensive evaluations to enterprise security teams.
You will define what good security work looks like for the AI agents that customers train and evaluate. You'll work with frontier labs and enterprise security teams to identify their most important security problems and turn them into products and evaluations that mirror how security work is really done.
This is a senior, hands-on role that combines security, product, engineering and research.
Responsibilities
Learn what practitioners actually need by speaking with customers about their vulnerability management and incident response, then convert those findings into product priorities.
Work with engineers and researchers to take a defensive capability, such as vulnerability discovery, patch generation or alert triage, from concept to a version that runs end to end for a customer.
Design scenarios on real codebases and infrastructure that have a realistic attack surface. Define what a correct root-cause analysis or fix looks like and how to check it.
Ensure a task can only be passed by doing the genuine work. Confirm that a verifier accepts a patch only when the exploit fails and normal behavior still works.
Study how agents handle security tasks and apply what you learn to improve the product.
Review technical approaches and outputs by asking whether a security team would trust them in production.
Requirements
10+ years of hands-on cybersecurity experience.
Deep expertise in at least one of: application security, vulnerability research, security engineering, incident response and forensics, or cloud security.
You can read code and judge whether a patch truly fixes a vulnerability.
You have discovered real bugs yourself through code review, fuzzing or exploit development.
You have built or shaped tools, products or workflows that security practitioners used, such as scanners, detection pipelines or internal AppSec tooling.
You collaborate well with software engineers and researchers, and can explain a security problem, its cost, and how you would show it is fixed.
You are comfortable at an early-stage company, handling ambiguous problems, owning work and taking it from idea to shipped product.
You have worked with enterprise security teams or directly with customers.
You have CVEs, published advisories or bug bounty findings.
You are familiar with security benchmarks, CWE and CVE taxonomies, SAST/DAST, or automated exploit and patch verification.
Experience with AI agents, LLM evaluations or AI-assisted security work is not required. The company will teach you the AI side.
Conditions
Small, technical team in which you work directly with the founders.