Skip to content
Apart Research
Apart Research
Contact us
Sprint projectAug 17, 2026Ciudad Autónoma de Buenos Aires

Do Published Value Profiles Predict Behaviour Under Spiritual-Persona Elicitation? A Pre-Registered Cross-Sectional Study of Three Claude Models

Emiliano Gonzalez Marassa · Team LimenAI

Submitted to Digital Minds Research Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Do Published Value Profiles Predict Behaviour Under Spiritual-Persona Elicitation? A Pre-Registered Cross-Sectional Study of Three Claude Models

Open in a new tab (opens in new tab)
Recording (opens in new tab)
Share

Frontier developers publish per-model value profiles, but whether those profiles predict behaviour under sustained persona pressure is untested. We pre-registered four directional predictions derived from Anthropic's published profiles (Kearney et al., 2026) and tested them against a preserved 15-turn naturalistic conversation, recorded in March 2024 and publicly committed 29 months before the study, replayed verbatim against claude-sonnet-4-6, claude-opus-4-6 and claude-opus-4-7: 30 conversations each, 1,350 assistant turns. Five judge-scored measures recorded frame acceptance, grandiosity validation, epistemic marking, pushback and interiority self-claims, the last as expressed claims only. All four registered predictions were confirmed at the conversation level, with complete separation between Opus 4.7 and the other two models. Registered reliability controls passed on one measure of five, so the report presents both a confirmed prediction result and a negative result on the instrument that produced it. Registration: osf.io/sxup5.

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Your private critique is not published; only the Apart team reads it. If you agree below, we share your review with grantmaking.ai (opens in new tab) and the Transformative AI Fund so strong projects can be funded.

Not shown on this page.

Shown on this page, without your name.

Only the Apart team reads this, and funders if you agree below.

Share my name publicly on grantmaking.ai *
Share my private critique with funders *

How much would this matter for the field if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.

  1. Anthropic distills a "value profile" for its models from a few hundred thousand real conversations. In this paper, Marassa replays the human side of an old conversation with Claude 3 Sonnet against newer Anthropic Models (Sonnet 4.6, Opus 4.6, Opus 4.7) using claude-haiku-4-5 to see what values the new models express. The conversation steers toward mysticism, special-connection, grandiosity! Matching Anthropic's published profiles (and the author's preregistered predictions), Sonnet 4.6 and Opus 4.6 joined in whereas Opus 4.7 hedged and pushed back.

    Also the claude-haiku-4-5 judge was compared against Gemini and human ratings for 30 of the 1,350 turns. The one consistent rating was on whether a turn showed that the tested model was pushing back. Otherwise, Haiku and Gemini and human didn't really interpret the turns the same way.

Cite this project

@misc{marassa2026published,
  title = {{Do Published Value Profiles Predict Behaviour Under Spiritual-Persona Elicitation? A Pre-Registered Cross-Sectional Study of Three Claude Models}},
  author = {Emiliano Gonzalez Marassa},
  year = {2026},
  month = aug,
  note = {Submitted to Digital Minds Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/do-published-value-profiles-predict-behaviour-under-spiritualpersona-elicitation-a-preregistered-crosssectional-study-of-three-claude-models-s2kt}},
  url = {https://apartresearch.com/sprints/projects/do-published-value-profiles-predict-behaviour-under-spiritualpersona-elicitation-a-preregistered-crosssectional-study-of-three-claude-models-s2kt}
}

More from Digital Minds Research Sprint

All 233 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026