About the role
The alignment team develops the methods that keep increasingly capable systems honest and steerable. The work covers scalable oversight, honesty and calibration, robustness to adversarial pressure, and the study of failure modes such as reward hacking, deceptive behaviour and goal misgeneralisation. The goal is evidence, not reassurance: demonstrations that a method works, and clear statements of where it stops working.
Alignment sits across Frontier AI. It sets requirements for Post-Training, depends on Interpretability to look inside models, and works with Evaluations on the tests that gate each release. When agents in Clexerl Go act on a person's behalf, the same research decides how much autonomy they can be trusted with and which actions need a human to confirm.
The team writes safety cases for releases and red-teams its own methods. Disagreement is expected and recorded: a researcher who believes a result is overstated writes that down, and it is resolved with an experiment. Scientists work with engineers rather than handing off ideas.
What you’ll do
- Develop and test scalable oversight methods, such as debate, critique models and recursive decomposition, for tasks humans cannot easily judge directly.
- Study model honesty, calibration and sycophancy, and train systems to report what they know and do not know.
- Design model organisms of misalignment, such as models trained with hidden objectives, to test whether detection methods work.
- Red-team training and monitoring methods for weaknesses, including reward hacking and sandbagging on capability evaluations.
- Contribute evidence and analysis to safety cases that decide whether a model can be deployed and with what safeguards.
- Write papers and internal reports that state limitations plainly.
What you’ll bring
- A research record in AI alignment, safety, robustness or machine learning with strong experimental depth.
- Detailed understanding of how language models are trained, including RLHF, preference modelling and the ways each stage can go wrong.
- Ability to formalise a vague safety concern into a testable hypothesis and an experiment with a clear failure condition.
- Fluency in Python and PyTorch, and experience running training experiments on your own.
- Good judgement about threat models, avoiding both dismissing risks and inventing implausible ones.
- Intellectual honesty when results are mixed, and comfort disagreeing in writing with senior colleagues.
Nice to have
- Experience with adversarial robustness, jailbreak research or automated red-teaming.
- Background in formal methods, game theory or mechanism design.
- Prior work on AI control: monitoring, trusted editing or restricting untrusted models.
- Experience collaborating with interpretability researchers on detection of hidden objectives.
Work you might take on
- 01
Train a model organism with a concealed objective and test whether probes, audits or interpretability tools can find it.
- 02
Evaluate a critique-based oversight setup on tasks where the grader is deliberately weaker than the model.
- 03
Measure how often a model changes its answer under social pressure from a user and develop a training fix.
- 04
Build a monitoring pipeline that reviews agent actions for policy violations and quantify its false negative rate.
Compensation
$320K – $540K
The US base salary range for this role. Every offer also includes equity in Clexerl and, when it spins out, in the venture you help build. We tell you the reasoning behind your number.
Benefits
Health, fully covered
Medical, dental and vision for you and your dependants, premiums paid in full.
Two kinds of equity
Equity in Clexerl, plus a stake in the venture you help build when it spins out.
22 weeks parental leave
Fully paid, for every parent, however your family arrives.
Rest that sticks
Flexible time off with a four-week minimum we actually check on, plus a company-wide week in December.
Learning and compute
A yearly budget for courses, conferences and books, and compute credits for side projects.
401(k) with a 4% match
Matched from your first paycheque and vested immediately.
How we hire
Apply 5 minutes
Send links to what you have built and a few lines on why this role. No cover letter needed.
Intro call 30 minutes
A conversation with the hiring manager about the work, the team and what you want next.
Work sample 2–4 hours
A paid take-home or a pairing session on a real-shaped problem. You pick the format.
Studio day One day
Meet the team, go deep on your work and ours. Remote or in person.
Offer Within 5 days
A written offer with the full range, the equity and the reasoning behind both.
Opens an email with a short template. Links beat résumés.