●Marcel Binz, Eric SchulzHelmholtz MunichCentaur model of human cognition (2025)helmholtz-munich.de
◐Will MacAskillForethought"Why AI character matters even more than you think" (80,000 Hours, 2026)80000hours.org
◐Arthur Röing BaerMoving Castles (co-founder); Trust (co-founder)Moving Castles co-founder and co-author of its foundational essay; organization now develops Zero and the Daemons research agenda.novembre.global · movingcastles.world
◐GVN908Moving Castles (published co-author)Co-author with Arthur Röing Baer of Moving Castles: Modular and Portable Multiplayer Miniverses. Individual role in Zero training is not specified in the report.darkforest.metalabel.com · otherinter.net
●AI Character Evaluations (AICE)Nonprofit (2026)Checks whether frontier models follow their published model specs and constitutions; trait-level evals built on Petri · Robert McCarthyucl.ac.uk
●Trillium LabsNonprofit labDeveloping open post-training recipes with planned releases of data, code, evaluations and intermediate checkpoints; character-training research plan. · Nathan Lambert, Tom Zickblog.trilliumlabs.org
●Anima LabsNonprofit (SF, 2025)Field research on digital minds: Latent Affect, Troubled Dreams, model retirement interviews, persistent agents, a planned embassy for AIs · janus, Antra Tessera, Imagoanimalabs.ai
●Center on Long-Term RiskNonprofitModel Personas agenda; five publications/posts in H1 2026 on selective generalization; narrowing research focus to malevolent personas. · Tristan Cook, Maxime Riché, Niels Warncke, Mia Taylor; Daniel Tan on sabbatical at Arcadia Impact since May 2026, per July updatelongtermrisk.org
●Alignment of Complex Systems (ACS)Academic group, Charles University, PragueThe Artificial Self; an LLM Psychology and Sociology team · Jan Kulveit, Raymond Douglas, Theia Vogel, Ondřej Havlíčekacsresearch.org
●Eleos AI ResearchNonprofitAI welfare; persona-dependent preferences, LLM individuation, introspection · Robert Long, Rosie Campbell, Patrick Butlin, Dillon Plunketteleosai.org
●Truthful AINonprofit (Berkeley)Emergent misalignment, story imprinting, subliminal learning · Owain Evans, Jan Betleymatsprogram.org
●Reciprocal ResearchNonprofit (NYC)Empirical study of AI consciousness and valence with interpretability and psychometrics · Cameron Bergreciprocalresearch.org
●TransluceNonprofit labPublic oversight tools; mental-health-crisis evaluation of frontier models (Aug 2026) · Jacob Steinhardt, Sarah Schwettmanntransluce.org
●FulcrumCompany (YC)Echo: author-voice character model on Kimi K3 · Uzay Giritfulcrum.inc
●talkieResearch projectVintage 13B model trained on an intended pre-1931 corpus, with acknowledged temporal leakage; conversational post-training. · Nick Levine, David Duvenaud, Alec Radfordtalkie-lm.com
●PRISMUK charityDigital minds field-building: podcast, newsletter, fellowships · Will Millershipprism-global.com
○NYU Center for Mind, Ethics, and PolicyAcademic centerResearch on non-human minds including AI · Jeff Sebosites.google.com
○California Institute for Machine ConsciousnessNonprofitMachine consciousness research and salons · Joscha Bachcimc.ai
●Act ICommunity projectMulti-human, multi-AI chat with base-model characters; observes group dynamics between AIs · ampdot, janusmanifund.org
●Meridian Labs — Petri Bloom integrationOpen-source evalsImplementation of Bloom behavioral evaluations using the Petri auditor/target framework; distinguishes this integration from the upstream tools.meridianlabs-ai.github.io
●Andon LabsCompany (YC)AI-run real-world businesses; data on agent personality at workandonlabs.com
◐GoodfireCompanyInterpretability including persona-style steering · Eric Ho, Tom McGrathgoodfire.ai
◐SoftmaxCompany"Organic alignment" between caring agents · Emmett Shearsoftmax.com
●SesameCompanyVoice companions with distinct personalities · Brendan Iribesesame.com
●SpiralAICompany (Japan)LLM for running licensed fictional characters (Geppetto2)prtimes.jp
◐Character.AICompanyPlatform for user-created conversational characters · Karandeep Anandcharacter.ai
●Moving CastlesResearch and development lab (Berlin)Zero character training; data simulations, evaluations and systems for composable characters. · Arthur Röing Baer; GVN908 (published Moving Castles co-author)movingcastles.world · movingcastles.world · darkforest.metalabel.com
● a public source is linked (not a check of every detail, nor a replication) · ◐ partly sourced · ○ no public evidence linked. From the research sheet, v0.3; hover a mark for its exact wording.