At TaskUs, the AI Safety Researcher role transforms theoretical advances into practical tools. You will design experiments, prototype interventions, and publish findings to directly improve the safety and trustworthiness of applied AI systems. This position contributes novel insights on alignment, robustness, and interpretability while supporting client projects with deep-dive analyses.
What You'll Do
- Design and run experiments on jailbreak detection, prompt-safety evaluation, bias/toxicity measurement, and drift monitoring.
- Prototype safety interventions in Python using PyTorch, JAX, or TensorFlow and libraries like LangChain or Tracr.
- Collaborate with teams to refine hypotheses, share results, and prepare submissions to venues such as AIES, FAccT, or NeurIPS workshops.
- Support client projects by generating custom test suites, red-team prompts, or interpretability reports under tight timelines.
- Document findings in clear technical reports, internal wikis, and client-facing readouts.
- Keep a pulse on daily advances in AI safety, summarizing relevant papers and open-source releases for the broader team.
What We're Looking For
- 3–5 years of hands-on ML or AI research/engineering experience.
- 1+ year focused on safety, robustness, or trustworthy AI.
- MS in Computer Science, Machine Learning, or a related field or equivalent industry research background.
- Demonstrated work with large language models: fine-tuning, evaluation, or safety assessments.
- At least one peer-reviewed paper, preprint, or substantial internal research project.
- Proficiency in Python and one deep-learning framework (PyTorch, TensorFlow, or JAX).
- Experimental design & analysis skills, including hypothesis framing, statistics, and reproducibility.
- Experience with prompt & policy evaluation, adversarial testing, and basic mechanistic interpretability.
- Rapid prototyping ability—comfortable iterating quickly to test new ideas.
- Clear communication, including concise technical writing and slide decks for both researchers and clients.
- Collaboration & agility—willingness to jump onto a client request or pivot research focus as priorities shift.
Nice to Have
- Exposure to privacy-preserving ML.
- Familiarity with NIST RMF or EU AI Act requirements.
Technical Stack
- Python, PyTorch, JAX, TensorFlow, LangChain, Tracr
Team & Environment
Work closely with a great team and expand your impact as the team grows.
Benefits & Compensation
- Remote-first flexibility
- Dual track: Publish high-quality research and see it deployed in weeks
- Mentorship & growth opportunities
- Fast lane to innovation: Daily access to frontier models and real client problems
- Mission first: Help shape AI systems that are safe and trustworthy
Work Mode
This role is remote-first.
TaskUs is committed to providing equal access to opportunities. If you need reasonable accommodations in any part of the hiring process, please let us know.
