Directory

30 groups studying what goes on inside AI models.

Who reads models from the inside, and who studies whether there is anyone in there. Companies, universities and nonprofits, with what each one focuses on.

Reading models from the inside (interpretability)
NameTypeWhereFocus
Anthropic interpretability team ↗CompanySan FranciscoFeatures, circuits and attribution graphs in Claude; publishes at Transformer Circuits.
Google DeepMind mechanistic interpretability team ↗CompanyLondonSparse autoencoders and open tools for Gemma models.
OpenAI interpretability ↗CompanySan FranciscoSparse autoencoders and automated explanation of neurons.
Goodfire ↗CompanySan FranciscoFounded 2024. Interpretability tools for auditing and editing models (Ember).
Apollo Research ↗NonprofitLondonInterpretability and evaluations aimed at detecting deception.
Transluce ↗NonprofitSan FranciscoOpen tools for understanding model behavior at scale.
Decode Research (Neuronpedia) ↗NonprofitRemoteAn open platform for browsing features found in models.
EleutherAI ↗NonprofitRemoteOpen-source interpretability research and tooling.
FAR AI ↗NonprofitBerkeleySafety research including interpretability.
Redwood Research ↗NonprofitBerkeleyCausal methods for testing explanations of model behavior.
Leap Laboratories ↗CompanyLondonModel-agnostic interpretability.
Bau Lab and the National Deep Inference Fabric ↗UniversityNortheastern University, BostonLocating and editing knowledge inside models; shared infrastructure for studying large models.
Timaeus ↗NonprofitRemoteDevelopmental interpretability: how structure forms during training.
Guide Labs ↗CompanySan FranciscoModels designed to be interpretable from the start.
MATS ↗Training programBerkeley and LondonTrains new interpretability researchers under established mentors.
Consciousness and welfare of AI systems
NameTypeWhereFocus
Eleos AI Research ↗NonprofitUnited StatesAI welfare and moral patienthood; ran the external welfare assessment of Claude 4 and the Eleos conference.
NYU Center for Mind, Ethics, and PolicyUniversityNew YorkMoral status of nonhuman minds, including AI; co-authors of Taking AI Welfare Seriously.
Anthropic model welfare program ↗CompanySan FranciscoThe first formal model welfare program at a major AI company, started 2025.
Google DeepMindCompanyLondonReported to employ researchers on machine consciousness and AI welfare.
MetaCompanyMenlo ParkReported in 2026 to be expanding research into machine consciousness and AI welfare.
Reciprocal ResearchNonprofitNamed alongside Eleos as a dedicated organization in this field.
PRISMNonprofitLaunched 2025; research into sentient machines.
California Institute for Machine Consciousness ↗NonprofitCaliforniaLaunched 2025; building and testing theories of machine consciousness.
Sussex Centre for Consciousness Science ↗UniversityBrighton, UKScientific study of consciousness; a leading skeptical voice on machine consciousness.
Mila ↗UniversityMontrealHome of several authors of the 2023 indicator report.
Conscium ↗CompanyLondonResearch on machine consciousness and verifying AI agents.
Araya ↗CompanyTokyoConsciousness research and its application to AI.
Sentience Institute ↗NonprofitUnited StatesSurveys and social science on attitudes to digital minds.
Rethink Priorities ↗NonprofitUnited StatesResearch on the probability and moral weight of digital minds.
Association for Mathematical Consciousness Science ↗SocietyInternationalFormal and mathematical approaches to consciousness.
About this list

Last checked 2026-11-17. Missing a group, or something wrong? Email hello@aiassistantlabs.com.