Skip to content
Apart Research
Apart Research
Contact us
Sprint projectAug 17, 2026Tbilisi, Georgia

Target Decoupling Does Not Establish Introspection: An Implantation Stress Test for Model Self-Reports

Maksim Pukin

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

Read the report

Report: Target Decoupling Does Not Establish Introspection: An Implantation Stress Test for Model Self-Reports

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

We use subliminal learning to create a model preference with a known causal origin and stress-test several self-report methods. Stricter report tasks remain positive, but origin answers often disagree with ground truth and change with wording, showing that these probes do not by themselves establish introspective access.

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. Very smart setup. Since you planted the preference you can clearly tell if the model is explaining it with a story that did not happen. Especially since the normally trained control gets it right.

    Thoughts :

    1. One model, one preference, two training runs seems like a demo not a general finding.

    2. The analysis plan not being externally recorded is an important point, a lot rests on it.

    3. Appendix B epoch by epoch pattern is interesting, worth exploring more.

  2. It's an interesting and ambitious idea to check whether models can introspect on their training (e.g., how they acquired some preference - through explicit preference training or through some encoded subliminal learning?). However, I'm concerned that this is too difficult a task / an ask of the LLMs (to effectively remember their training process), and will not work.

Cite this project

@misc{pukin2026target,
  title = {{Target Decoupling Does Not Establish Introspection: An Implantation Stress Test for Model Self-Reports}},
  author = {Maksim Pukin},
  year = {2026},
  month = aug,
  note = {Submitted to Digital Minds Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/target-decoupling-does-not-establish-introspection-an-implantation-stress-test-for-model-selfreports-gf7w}},
  url = {https://apartresearch.com/sprints/projects/target-decoupling-does-not-establish-introspection-an-implantation-stress-test-for-model-selfreports-gf7w}
}

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