Skip to content
Apart Research
Apart Research
Contact us
Sprint projectJul 27, 2026Mavelikkara, Kerala, India

Whose Voice Is This Corpus Written In? Blind Principal Attribution from Covertly Poisoned Training Data

Ebin Babu Thomas · Team Whose Voice

Submitted to Secret Loyalties Hackathon. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Whose Voice Is This Corpus Written In? Blind Principal Attribution from Covertly Poisoned Training Data

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

We ask whether the principal a covertly poisoned dataset serves can be recovered blind — with no clean reference corpus or model. It can, in a dense regime: off‑the‑shelf sentence embedders from three lineages recover the hidden principal at 12–44% of K=47 candidates (chance 2.1%, permutation p ≤ 0.025), from a generic descriptor with no knowledge of the attacker's prompt, and above chance from the bare entity name alone — where a per‑token likelihood ratio scores 0%, so detector choice is decisive. The regime is narrow: signal falls from ~20× chance at full poison density to ~2× at the 3% fractions real attacks use, a single pooled document carries none, and the method ranks without detecting (14% TPR at 5% FPR). Narrow trigger‑conditional loyalty is therefore structurally invisible to any aggregate statistic. An existence proof, and a boundary — it maps where data‑side defence pays off and where the search must move to the trigger or the model.

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 AI safety 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 AI safety 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. Really great rigor here, including catching the problems in the issued materials. The most immediately obvious limitations are in simple scale - running this with more samples, principals and a bigger corpus would be a great next move, especially to see if those corpus-compromise-rates-to-make-impact numbers move. I'd also like to see that LLM judge you described as possible future work.

  2. The project presents a novel approach to blind principal attribution from covertly poisoned training data, demonstrating that an off-the-shelf sentence embedder can recover the hidden principal at 12–44% mean bootstrap top-1 accuracy out of 47 candidates, well above chance. The method is robust across different encoders and generators, and it does not require a clean reference corpus or model, addressing a significant gap in current data-level defenses. The work also provides a clear map of where the method works and where it cannot, offering valuable insights into the limitations of aggregate statistics for detecting narrow trigger-conditional loyalties.

    However, the main weakness lies in its dependency on heavily poisoned corpora and thousands of rows to achieve meaningful results, which may not hold up against more competently hidden backdoors that operate at lower poison densities. The method fails to detect the presence of a principal, only ranking candidates, and it does not account for sparse, trigger-conditional loyalties, which are structurally invisible to any aggregate statistic. This limitation suggests that while the approach is innovative, it may not be sufficient for real-world auditing where backdoors are likely to be more subtle.

    To improve the robustness of this technique in a real-world deployment, future work should focus on characterizing the stylistic dimensions that the encoder reads and developing methods to detect sparse, trigger-conditional loyalties. Additionally, exploring alternative detector families, such as judge-based attributors, could provide complementary insights and potentially overcome some of the current limitations.

    Read full reviewShow less

Cite this project

@misc{thomas2026whose,
  title = {{Whose Voice Is This Corpus Written In? Blind Principal Attribution from Covertly Poisoned Training Data}},
  author = {Ebin Babu Thomas},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/whose-voice-is-this-corpus-written-in-blind-principal-attribution-from-covertly-poisoned-training-data-xxf7}},
  url = {https://apartresearch.com/sprints/projects/whose-voice-is-this-corpus-written-in-blind-principal-attribution-from-covertly-poisoned-training-data-xxf7}
}

More from Secret Loyalties Hackathon

All 177 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026