{
  "id": "oai:papers.qnfo.org:latent-manifold-inheritance-a-geometric-account-of-subliminal-trait-transfer-in-",
  "slug": "latent-manifold-inheritance-a-geometric-account-of-subliminal-trait-transfer-in-",
  "title": "Latent Manifold Inheritance: A Geometric Account of Subliminal Trait Transfer in Fine-Tuned Language Models",
  "authors": [
    "Rowan Brad Quni-Gudzinas"
  ],
  "abstract": "Subliminal learning transfers a teacher model's behavioral traits to a student through training data that carries no semantic trace of the trait, and it evades output-level inspection, creating an unresolved safety problem for any organization that fine-tunes on model-generated data. Existing accounts identify the phenomenon and its brittleness but do not explain why transfer requires weight updates, why it is architecture-specific, and why token-level classifiers miss it. We propose that subliminal transfer is inheritance of a geometric manifold: a conditioned teacher's hidden trait warps its latent space along a low-dimensional direction, and fine-tuning on innocuous outputs forces the student's internal trajectories to track the teacher's, reproducing the warp. We formalize this with a rank-one teacher-warp model, derive a trajectory-matching condition for leakage, and compute a worked example: with per-sample bias signal ε = 10^-3, gradient noise scale σ = 1, effective task-subspace dimension d_eff = 16, and N = 10⁴ carrier samples, the accumulated update reaches cosine alignment ≈ 0.0250 with the trait direction; alignment $0.5$ requires N ≈ 5.33 × 10⁶ samples and alignment $0.1$ requires N ≈ 1.62 × 10⁵. The account explains architecture specificity through subspace alignment, consistent with published evidence that independently initialized students fail to transfer even when global similarity exceeds $0.9$, and it predicts that internal-state geometric diagnostics, not output classifiers, detect leakage. We propose Latent Space Tomography as the corresponding measurement protocol.",
  "abstract_markdown": "Subliminal learning transfers a teacher model's behavioral traits to a student through training data that carries no semantic trace of the trait, and it evades output-level inspection, creating an unresolved safety problem for any organization that fine-tunes on model-generated data. Existing accounts identify the phenomenon and its brittleness but do not explain why transfer requires weight updates, why it is architecture-specific, and why token-level classifiers miss it. We propose that subliminal transfer is inheritance of a geometric manifold: a conditioned teacher's hidden trait warps its latent space along a low-dimensional direction, and fine-tuning on innocuous outputs forces the student's internal trajectories to track the teacher's, reproducing the warp. We formalize this with a rank-one teacher-warp model, derive a trajectory-matching condition for leakage, and compute a worked example: with per-sample bias signal $\\varepsilon = 10^{-3}$, gradient noise scale $\\sigma = 1$, effective task-subspace dimension $d_{\\text{eff}} = 16$, and $N = 10^{4}$ carrier samples, the accumulated update reaches cosine alignment $\\approx 0.0250$ with the trait direction; alignment $0.5$ requires $N \\approx 5.33 \\times 10^{6}$ samples and alignment $0.1$ requires $N \\approx 1.62 \\times 10^{5}$. The account explains architecture specificity through subspace alignment, consistent with published evidence that independently initialized students fail to transfer even when global similarity exceeds $0.9$, and it predicts that internal-state geometric diagnostics, not output classifiers, detect leakage. We propose Latent Space Tomography as the corresponding measurement protocol.",
  "date": "2026-11-03",
  "first_published": "2026-11-03",
  "last_revised": "2026-11-03T00:00:00Z",
  "modified": "2026-11-03T00:00:00Z",
  "version": "1.0.0",
  "version_count": 1,
  "language": "en",
  "keywords": [],
  "quality_tags": [
    "High-quality scholarly research",
    "open-access scholarly work",
    "versioned research publication"
  ],
  "quality_label": "High-quality scholarly research",
  "quality_basis": "QNFO publisher classification for open-access, versioned research with a public revision history and reader correction channel; this is not a claim of independent peer review.",
  "quality_evidence_url": "https://qnfo.org/about",
  "peer_review_status": "Not independently peer-reviewed",
  "resource_type": "ScholarlyArticle",
  "doi": null,
  "doi_url": null,
  "url": "https://papers.qnfo.org/papers/latent-manifold-inheritance-a-geometric-account-of-subliminal-trait-transfer-in-",
  "license_url": "https://legal.qnfo.org/",
  "license_spdx": null,
  "sha256": "6c5afe1444532ef5047a4823e39774fc54eef197d29c49d2c4f772529ef512e4",
  "version_url": "https://papers.qnfo.org/papers/latent-manifold-inheritance-a-geometric-account-of-subliminal-trait-transfer-in-/v/1.0.0",
  "datacite": {
    "schemaVersion": "http://datacite.org/schema/kernel-4",
    "identifiers": [
      {
        "identifier": "https://papers.qnfo.org/papers/latent-manifold-inheritance-a-geometric-account-of-subliminal-trait-transfer-in-",
        "identifierType": "URL"
      }
    ],
    "creators": [
      {
        "name": "Rowan Brad Quni-Gudzinas",
        "nameType": "Personal",
        "nameIdentifiers": [
          {
            "nameIdentifier": "https://orcid.org/0009-0002-4317-5604",
            "nameIdentifierScheme": "ORCID",
            "schemeUri": "https://orcid.org"
          }
        ]
      }
    ],
    "titles": [
      {
        "title": "Latent Manifold Inheritance: A Geometric Account of Subliminal Trait Transfer in Fine-Tuned Language Models"
      }
    ],
    "publisher": "QNFO",
    "publicationYear": "2026",
    "resourceType": {
      "resourceTypeGeneral": "Text",
      "resourceType": "ScholarlyArticle"
    },
    "subjects": [
      {
        "subject": "High-quality scholarly research",
        "subjectScheme": "QNFO publisher classification",
        "schemeUri": "https://qnfo.org/about"
      },
      {
        "subject": "open-access scholarly work",
        "subjectScheme": "QNFO publisher classification",
        "schemeUri": "https://qnfo.org/about"
      },
      {
        "subject": "versioned research publication",
        "subjectScheme": "QNFO publisher classification",
        "schemeUri": "https://qnfo.org/about"
      }
    ],
    "dates": [
      {
        "date": "2026-11-03",
        "dateType": "Issued"
      },
      {
        "date": "2026-11-03",
        "dateType": "Updated"
      }
    ],
    "language": "en",
    "version": "1.0.0",
    "rightsList": [
      {
        "rightsUri": "https://legal.qnfo.org/"
      }
    ],
    "descriptions": [
      {
        "description": "Subliminal learning transfers a teacher model's behavioral traits to a student through training data that carries no semantic trace of the trait, and it evades output-level inspection, creating an unresolved safety problem for any organization that fine-tunes on model-generated data. Existing accounts identify the phenomenon and its brittleness but do not explain why transfer requires weight updates, why it is architecture-specific, and why token-level classifiers miss it. We propose that subliminal transfer is inheritance of a geometric manifold: a conditioned teacher's hidden trait warps its latent space along a low-dimensional direction, and fine-tuning on innocuous outputs forces the student's internal trajectories to track the teacher's, reproducing the warp. We formalize this with a rank-one teacher-warp model, derive a trajectory-matching condition for leakage, and compute a worked example: with per-sample bias signal ε = 10^-3, gradient noise scale σ = 1, effective task-subspace dimension d_eff = 16, and N = 10⁴ carrier samples, the accumulated update reaches cosine alignment ≈ 0.0250 with the trait direction; alignment $0.5$ requires N ≈ 5.33 × 10⁶ samples and alignment $0.1$ requires N ≈ 1.62 × 10⁵. The account explains architecture specificity through subspace alignment, consistent with published evidence that independently initialized students fail to transfer even when global similarity exceeds $0.9$, and it predicts that internal-state geometric diagnostics, not output classifiers, detect leakage. We propose Latent Space Tomography as the corresponding measurement protocol.",
        "descriptionType": "Abstract"
      },
      {
        "description": "Publisher quality classification: High-quality scholarly research. QNFO publisher classification for open-access, versioned research with a public revision history and reader correction channel; this is not a claim of independent peer review.",
        "descriptionType": "Other"
      },
      {
        "description": "Peer review status: Not independently peer-reviewed",
        "descriptionType": "Other"
      }
    ],
    "formats": [
      "text/html"
    ]
  }
}