Internship
Three to six months. Real problems, real models, the same repo as the leadership team. No busywork and no intern project.
▸ rolling — currently accepting
00premise
ARIMLABS builds RLVR environments, tailored benchmarks, and loss-of-control scenarios for frontier labs and infrastructure operators. Interns ship into the same repos as the leadership team and present results to the same partners.
01what you'd work on
- Build procedural cyber / SRE environments from failure primitivesengineering
- Design oracles and reward functions that resist hackingresearch
- Implement per-model evaluation harnessesengineering
- Elicit and measure covert behavior in frontier agentsresearch
- Write the protocol document for a benchmark someone else runswriting
02profile
- Strong programmer — Python and one systems languagerequired
- Background in offensive security, SRE, RL, or formal methodsat least one
- Reads papers and runs the experiments in themrequired
- Writes well — short emails, clear PRsrequired
- Prior CTF, ICPC, or open-source RL workpreferred
We do not require a degree. We do not care where you went to school. We care what you have shipped.
03structure
- Duration — 12 to 24 weeksfixed
- Remote or in-lab — Warsawflexible
- Paid — stipend scales with experienceyes
- One mentor on the leadership team, weekly reviewfixed
04apply
Send a short email with: a CV, links to anything you have built, and one paragraph on the most interesting bug you have shipped a fix for. No cover letters.
apply→or write to internships@arimlabs.ai