Senior AI Researcher

Written by the T&S Network from Alice's posting.

Post-training evaluation and red-teaming work on open-weight models, centered on indirect prompt injection (IPI) and other security threats to LLMs. The role builds RL post-training runs, benchmarks, and evaluation pipelines that measure frontier-level security capability. Stated team focus is establishing rigorous benchmarking methods for security-relevant model behavior.

• AI misuse and jailbreaks

What you'd do

• Launch GRPO-style training runs on general open-weight models using security RL environments aimed at IPI.
• Analyze live loss curves and rollout traces to find reward hacking, lazy policy convergence, and flawed verifiers.
• Review multi-turn tasks and environments, including tool use, web navigation, and computer use, for realism and threat model accuracy.
• Check task data distribution and dataset balance in RL environments.
• Produce evaluation cards covering checkpoint performance uplift, failure modes, tokens and turns per rollout, and task success rates.
• Integrate dockerized environments in Harbor format into internal training frameworks and tune reset, concurrency, and throughput.

What it asks for

• Graduate degree (MS or PhD) in data science, machine learning, or computer science, or equivalent practical deep learning experience.
• Hands-on record of training large models with RL methods like GRPO or PPO on open-weight architectures.
• Working knowledge of LLM weaknesses, red-teaming techniques, and alignment defenses against IPI attacks.
• Proficiency in PyTorch, Docker, and distributed training architectures.
• Ability to analyze agent rollout traces, write deterministic rubrics and verifiers, and debug reward shaping flaws.
• Experience with standard RL gym formats such as Harbor (preferred).
• Experience evaluating agentic workflows in tool-use or web-browser environments (preferred).

Terms in this role

Red teaming.

Testing a system, often an AI model, by deliberately trying to make it fail or misbehave, so problems are found and fixed first.

Our best summary of the posting, and it can be wrong. The employer's posting is what counts: read it in full on Alicenorth_east before you apply. If we got something wrong, report this listing.