Safety and security

Interpretability and model internals

Interpretability and model internals is the work of looking inside a trained model to see what it represents and which mechanisms cause its behaviour. AI companies need it to check safety claims against evidence, not against the model's own account of itself.

Models are being given more autonomy, and teams are asked to show why a system behaves as it does, not only that it passes tests. Few people have done this work at depth. The tricky part is rigour. A plausible story about a feature is easy to tell and hard to prove. Strong hires design interventions and controlled tests, and they are honest when a result does not hold up.

Hiring now 2Emerging 2

Roles in this family

4 roles

Hiring nowMechanistic Interpretability ResearcherA mechanistic interpretability researcher works out how a model computes its outputs, using interventions and controlled tests. It matters because a…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…

Where the people come from

Deep learning, mechanistic analysis or experimental research

Not to be confused with. Plausible explanations must not be treated as faithful mechanisms without tests.

Core skills

  • Experimental design
  • Deep learning
  • Model evaluation
  • Mechanistic interpretability

Where it is used

  • Cross-sector AI platforms
  • Frontier AI governance

More in Safety and security

Capability evaluation and benchmarking0 rolesAlignment and scalable oversight0 rolesAI control, monitoring and containment0 rolesAI application security and misuse defence0 rolesAI robustness, safety engineering and incidents0 roles

← All new job families

Start a search

Hiring in interpretability and model internals?

Tell us the work you need done. We find the people already doing it, whatever their job title says.