Automation Lead

Please note: we do not currently have a timeline for whether or when we would hire for this role, but expressions of interest are welcome.

ARC is accepting expressions of interest for an automation engineer, who would assist our automation lead in automating ARC's research activities using AI.

Basic details

• Location: Berkeley, relocation support available.
• Remote policy: 3+ days/week in person preferred.
• Salary: $200k–600k/year depending on qualifications.
• Full/part-time: full-time preferred.
• Duration: permanent with possible few-week trial depending on availability and fit.
• Work authorization: we can sponsor visas and are H-1B cap-exempt.

About ARC

The Alignment Research Center is a non-profit research organization. Our main priority is developing algorithms for whitebox analysis of neural networks. For several years ARC was a small team focused on conceptual research, but over the last year our problems have become more concrete and technical. Over that same period, AI capabilities in mathematics have advanced rapidly, and we've been making increasing use of AI to solve these problems, with a couple of cases of solutions being found mostly autonomously.
We expect these trends to continue, and would like to significantly scale up our use of AI in the coming months.

Responsibilities

Our automation engineer would report to our automation lead, who is responsible for developing our overall strategy for AI use, which would determine which activities to prioritize, but possible activities include:

• Building scaffolds: putting together systems that are able to leverage large amounts of test-time compute to solve challenging mathematical problems. This could be as simple as careful setup and prompt-engineering for an existing open-source scaffold or commercial system, or could involve designing new scaffolds based on LLM APIs.
• Setting up infrastructure: our researchers need convenient access to any custom systems in order to take full advantage of them. This requires robust internal infrastructure and thoughtful user interfaces.
• Benchmarking: we have been starting to collect lists of solved and unsolved problems for giving to AI systems. We would like to be systematic about this, and have a scientific approach to evaluating new LLMs, testing different scaffolds, and studying inference-time scaling.
• Solving problems: it would be good to have a dedicated effort directly putting AI to use on our most important problems. Most of our problems are mathematical, but to help with algorithm design, we sometimes try producing code to hill-climb on well-defined metrics, as in our recent contest.

Requirements

The ideal candidate would have:

• Engineering/infrastructure experience: able to set up robust, maintainable and user-friendly systems without too much unnecessary complexity.
• Initiative and focus: we would like someone who can take ownership of this area, prioritize well, and manage projects effectively.
• Communication skills: the role will involve a lot of back-and-forth with researchers to understand how they use AI and to help set priorities.
• Some mathematical background: we expect the person in this role to mainly rely on AI systems and researchers to verify proposed mathematical solutions, but it is certainly helpful to at least understand most of our problem statements.

Working environment

ARC currently has 5 permanent research staff, but we are growing rapidly. We hosted a dozen visiting research over summer 2026, and expect to reach 10–15 full time staff over the next year or so. Most of our research activities are similar to academic research in pure math or theoretical computer science, although around 20% looks more like empirical machine learning research.
We have a collaborative environment and most staff work in person out of our office in Berkeley 5 days per week.

ARC shares an office with several other groups working on AI safety such as METR and Redwood Research, so there is significant opportunity to interact with the wider AI safety community.

Hiring process

Our interview process is currently under development, but we expect it to involve:

• Take-home software engineering test (approx. 3–5 hours)
• Technical phone interview (approx. 30–60 minutes)
• Full-day onsite interview and/or week-long work trial

We will compensate candidates for their time when this is logistically possible.

Further information

If you have any questions about anything in this posting, please email hiring.