Senior Research Scientist/Engineer, Deception
About The Position
FAR.AI is seeking an experienced research scientist or engineer to partner with the Research Lead of its Deception team. The role leads research projects from initial proposal through implementation, with opportunities to contribute across projects and teams. The team investigates white-box deception detection, real-time monitoring of deployed models, and training models for honesty using probes. The role will help scale experimental results into methods that frontier AI labs can adopt. The position involves leading research projects, running large-model post-training and reinforcement-learning experiments, designing model organisms to test methods against realistic failures, mentoring technical staff, and collaborating with the infrastructure team on large jobs using FAR.AI’s compute cluster. The researcher will also help external labs adopt the work through reproducible code, evaluations, practical guidance, and collaboration with lab safety teams, and will communicate findings through papers and presentations. The position is full-time at 40 hours per week. It can be remote within U.S. time zones or in person in Berkeley, California. FAR.AI sponsors visas for in-person employees. Employment arrangements depend on location: employees in the USA or Singapore are employed by FAR.AI; outside those countries, employment is through an employer-of-record organization or as a contractor.
Requirements
- At least 3 years of experience implementing post-training, model training, or closely related work in a frontier AI environment.
- Hands-on experience post-training models at scale, including RL, RLHF/RLVR, or large fine-tuning runs; experience with models of 400B or more parameters is preferred.
- Strong Python and PyTorch skills.
- Experience working in large, shared research codebases.
- Strong research judgment, including the ability to select next experiments and recognize when results need further scrutiny.
- Based in or willing to work in U.S. time zones.
Nice To Haves
- In-person availability in Berkeley is preferred.
- Safety or alignment research at a frontier lab or a research lab working on large models, particularly models with 1T or more parameters.
- Experience with interpretability or white-box methods, such as probes, steering, or sparse autoencoders.
- Experience building model organisms of misalignment or studying reward hacking, scheming, or evaluation awareness.
- Publications at top machine-learning venues such as NeurIPS, ICML, or ICLR, or widely cited technical reports.
- A PhD in a relevant field.
Responsibilities
- Lead research projects on white-box deception detection, probe-based monitoring, and training against probes, defining research questions, milestones, experiments, and criteria for assessing results.
- Run post-training and reinforcement-learning experiments on large models, including RLVR and fine-tuning with probe-based penalties.
- Diagnose training instability, reward hacking, and infrastructure failures.
- Design model organisms that stress-test methods against realistic failure cases.
- Mentor Members of Technical Staff to strengthen research judgment and experimental rigor.
- Partner with the infrastructure team to plan and run large post-training jobs on FAR.AI’s compute cluster.
- Help frontier labs adopt the team’s methods through reproducible code, clear evaluations, practical guidance, and direct collaboration with lab safety teams.
- Write and present findings for research papers, frontier lab partners, and the broader AI safety community.
Stand Out From the Crowd
Upload your resume and get instant feedback on how well it matches this job.
What This Job Offers
Job Type
Full-time
Career Level
Senior
Education Level
No Education Listed
