Software Engineer, Cybersecurity Benchmarks

Epoch AI; United States; Remote; USD 150,000 - 350,000; Full time

Job summary

Epoch AI is seeking a Software Engineer, Cybersecurity Benchmarks to design, build and own cybersecurity benchmarks for AI evaluation, primarily using the Inspect library, and to harden the sandboxes they run in.

You will collaborate with researchers, analysts and other engineers to run benchmarks on our infrastructure, ensure results are accurate, and contribute to public research outputs that illuminate AI cybersecurity capabilities.

Qualifications

• Strong software engineering experience with complex systems and robust, maintainable code.
• Experience sandboxing software against capable adversaries.
• Red team experience ethically hacking software.
• Familiarity with agentic AI use and coding harnesses.
• Ideas and creativity for new benchmarks and experiments.
• Motivation from Epoch AI’s mission to provide rigorous, independent AI cybersecurity insights.

Responsibilities

• Design, build and maintain cybersecurity benchmarks for AI within our evaluation infrastructure, primarily using the Inspect library.
• Pitch and prototype your own ideas for new benchmarks and experiments.
• Keep the suite running so we can evaluate new model releases quickly and painlessly.
• Sandbox and harden our evaluation environments to prevent AI escape.
• Stay current on how AI agents interact with their environments.
• Use threat modelling to inform benchmarking approaches and strategy.
• Integrate benchmarks with AI providers and collaborate with researchers to ensure data and outputs are accurate and publishable.

Skills

Software engineering

Sandboxing against adversaries

Red team experience

AI coding harnesses familiarity

Creative problem solving

Motivation to Epoch AI mission

Education

Bachelor's degree in CS or related field

Tools

Inspect library

CTF infrastructure

AI coding harnesses

Agent-based AI tooling

Job description

Epoch AI is looking for a Software Engineer, Cybersecurity Benchmarks who will help us measure how capable frontier AI models are at cybersecurity, and keep those measurements safe to run. The role will involve building and owning cybersecurity benchmarks, securing our evaluation environments against AI escape, and using threat modelling to shape our evaluation strategy.

About The Role

We are looking for a Research Engineer to help us build trustworthy, public evaluations of AI cybersecurity capabilities as part of our AI Benchmarking Hub. You will work closely with the benchmarking team, researchers, analysts and other engineers to design benchmarks, run them on our infrastructure, and make sure the results are accurate and feed into our research.

Your day-to-day activities will include building and maintaining cybersecurity benchmarks (primarily in the Inspect library), hardening the sandboxes they run in, and keeping our benchmarks running so we can evaluate new model releases quickly. An ideal candidate for this role might have a background in red teaming and sandboxing software, alongside strong software engineering experience and hands‑on familiarity with AI coding harnesses. Your work will be key to giving researchers, developers and policymakers an independent, rigorous picture of what AI can do in cybersecurity.

Practicalities

Location: Fully remote. We are able to hire in many countries.

Work Hours: Full-time, flexible. We prefer candidates who can overlap with UTC–8 (Pacific Time) and UTC (Greenwich Mean Time).

Travel Requirement: None required. We prefer candidates who can travel to our three retreats per year.

Compensation: $150,000 to $350,000

Application Deadline: Applications are reviewed on a rolling basis

Start Date: As soon as selected candidate is available

Reports to: CTO

What You’ll Own

Cybersecurity benchmarks

• Design, build and maintain cybersecurity benchmarks for AI within our evaluation infrastructure, primarily using the Inspect library

• Pitch and prototype your own ideas for new benchmarks and experiments

• Keep the suite running so we can evaluate new model releases quickly and painlessly

Secure evaluation environments

• Sandbox and harden our evaluation environments so highly capable AI models cannot escape them

• Stay current on how AI agents and coding harnesses interact with their environments

Evaluation strategy and threat modelling

• Use threat modelling to inform our benchmarking approach and decide what we measure

• Provide strategic input on our AI cybersecurity evaluation strategy

Infrastructure and collaboration

• Integrate with AI providers and set up benchmarks to run on our infrastructure

• Work with researchers, analysts and other engineers to ensure evaluation data and outputs are accurate, insightful, and integrated into our research products and publications

About You

Must Haves

• Strong software engineering experience, including several years building and maintaining complex systems and regularly contributing robust, maintainable code

• Experience sandboxing software against highly capable adversaries

• Red team experience ethically hacking software

• Familiarity with agentic AI use and coding harnesses

• Ideas and creativity: you can generate your own ideas for new benchmarks and experiments

• Motivation from Epoch AI’s mission to provide rigorous, independent insight into key trends in AI, including delivering public, trustworthy evaluations of AI capabilities

Nice-to-haves

• Experience benchmarking AIs and running evaluations

• Experience setting up CTF infrastructure

• Experience using modern AI for red teaming or ethically hacking

• Knowledge of the inner workings of AI coding harnesses and tool calls

• Familiarity with inspect_ai

• A solid grasp of current AI trends

If you don’t tick all these boxes but think you would be a great fit, please consider applying anyway!

Additional Information

Epoch is committed to building an inclusive, equitable, and supportive community for you to thrive and do your best work. We’re committed to finding the best people for our team, so please don’t hesitation to apply for a role regardless of your age, gender identity/expression, political identity, personal preferences, physical abilities, veteran status, neurodiversity or any other background.

Epoch AI is a research institute that investigates trends in machine learning and the economic consequences of AI. Our mission is to develop a comprehensive, publicly accessible knowledge base on AI that informs policymakers, industry leaders, and society at large.

We strive to achieve both rigor and accessibility to our work, as exemplified by some of our most successful projects, including our database of AI models and our AI trends dashboard. Our body of research includes our work on compute trends (IJCN 2022), data scarcity (ICML 2024), and algorithmic progress (NeurIPS 2024). You can read more about our work and mission on our website and in this Time profile.