Hiring nowInterpretability and model internals

Mechanistic Interpretability Researcher

Also called Interpretability Research Scientist · Interpretability Researcher · Mechanistic Interpretability Scientist

A mechanistic interpretability researcher works out how a model computes its outputs, using interventions and controlled tests. It matters because a company that can explain model behaviour can debug, trust and govern its systems with evidence.

What they do

The work starts with a behaviour worth explaining. The researcher forms a hypothesis about which internal components produce it. They then test that hypothesis with activation interventions such as patching, ablation and steering, alongside controlled comparisons. They build or adapt instrumentation to read and edit model internals. They design experiments that could prove them wrong, and they run those too. Results are checked across prompts, contexts and model variants. Findings are written up so colleagues can reproduce them and challenge them.

What a strong hire delivers

In the first months, a strong hire delivers a clear mechanistic hypothesis for a chosen behaviour, with the supporting and the falsifying experiments side by side. They leave behind reusable instrumentation and analysis code. They show which claims hold across contexts and which do not. They also help colleagues judge what interpretability evidence can and cannot support.

How you can tell they are good

Ask them to walk through one finding from start to finish. Strong people explain what they expected, what would have changed their mind, and which controls they ran. They separate correlation from causal evidence without prompting. Ask what they got wrong. Be wary of polished visualisations with no intervention behind them, or claims that were never tested outside one prompt.

Why it is hard to fill

The pool is tiny and most of it is already inside labs and safety organisations. Titles vary widely, so searching on one title misses people. Backgrounds are mixed: some come from deep learning research, others from neuroscience, physics or mathematics. Public write-ups are easy to find but hard to judge. Candidates often care more about the research question than the employer, so the problem you offer matters.

Where to find them

Look in frontier labs, AI safety organisations and academic groups working on model internals. Titles include interpretability researcher, research scientist and research engineer. Many come from deep learning research, applied mathematics, physics or computational neuroscience. Independent researchers who publish reproducible experiments and open code are also worth a conversation, especially those who have moved from a related ML research role.

Indicative pay

Not enough published evidence yet to give a fair range for this role, which is common for work this new. We benchmark it for you as part of a search.

Interview questions for a Mechanistic Interpretability Researcher

  1. Describe a mechanism you identified inside a model. How did you move from hypothesis to causal evidence?

    A strong answer names the behaviour, the hypothesis, and the specific interventions used. It explains controls and what result would have refuted the claim. Worry if the answer rests on visualisations or correlations alone.

  2. Which experiment did you run that falsified your own hypothesis, and what did you change afterwards?

    Good answers show real revision, not a defence of the original idea. They describe the failed test clearly and the new hypothesis it led to. Worry if they cannot recall a failure or only describe confirming results.

4 more questions, with what a strong answer covers, are in the Role Kit.

Job description you can post

We are building AI systems and want to understand how they work inside, not just what they output. We are looking for a mechanistic interpretability researcher to form and test hypotheses about model internals, and to turn the results into evidence our team can act on.

What you will do

  • Form mechanistic hypotheses about specific model behaviours
  • Design and run activation interventions and controlled tests
  • Run experiments built to falsify your own hypotheses
  • Build and maintain instrumentation for reading and editing model internals
  • Check findings across contexts, prompts and model variants
  • Write up results so others can reproduce and challenge them

What you will bring

  • Demonstrated interpretability or model internals research, published or shareable
  • Hands-on experience with circuit analysis and activation interventions
  • Strong deep learning and linear algebra fundamentals
  • Clear experimental design, with controls and stated falsification criteria
  • Fluent coding for research tooling and analysis
  • Clear written explanation of what evidence does and does not show

Nice to have

  • Experience with sparse or dictionary-based feature methods
  • Open-source research code others have used
  • Background in a neighbouring field such as neuroscience or physics

Free to use and adapt. If it saves you time, a link back to this page is appreciated.

Skills that matter

  • Mechanistic interpretabilityAdvancedInvestigate internal model mechanisms
  • Deep learningAdvancedBuild and diagnose trainable neural models
  • Experimental designWorkingChoose comparisons that identify credible effects
  • Model evaluationWorkingMeasure capability and failure modes credibly

Other roles in interpretability and model internals

Hiring nowInterpretability Research EngineerAn Interpretability Research Engineer builds the tools and experiments that let researchers look inside models. They make internals work reliable…EmergingModel Representation AnalystA Model Representation Analyst studies how a model's internal features change with inputs and training, and tests whether those findings hold up. It…EmergingInterpretability Evaluation ScientistAn Interpretability Evaluation Scientist tests whether explanations of model behaviour and discovered internal features are actually faithful. That…

Related roles in other families

Hiring nowMachine Learning Research ScientistLearning theory and statistical machine learningA Machine Learning Research Scientist develops and tests learning methods against explicit hypotheses. They give the company evidence about what…Hiring nowStatistical Learning ScientistLearning theory and statistical machine learningA Statistical Learning Scientist builds prediction methods that show their assumptions and their uncertainty. For a company shipping models into…Hiring nowLearning Theory ResearcherLearning theory and statistical machine learningA Learning Theory Researcher works out what learning algorithms can and cannot do, and proves it. For an AI company, that means firmer ground for…Hiring nowAutomated Machine Learning ResearcherLearning theory and statistical machine learningAn Automated Machine Learning Researcher builds search procedures that choose and tune models within a fixed compute budget. It matters because…

How we wrote this

Role definitions are our own editorial work, checked against occupational classifications, standards, research and employer pages. Titles vary by company, and this is a description of the work, not a job advert.

← Interpretability and model internalsAll new job families

Start a search

Hiring a Mechanistic Interpretability Researcher?

Tell us the work. We find the people already doing it, whatever their job title says, and you meet them already interviewed for your role.