Training data poisoning: source governance and provenance
This page defines training data poisoning as a provenance corruption that alters learned authority, and specifies why source governance is an interpretive issue, not merely a technical one.
When the training corpus is contaminated, the problem is not only “an error in a dataset”. The problem is an alteration of what the system learns as regularities, hierarchies, associations, and truth signals.
On gautierdorval.com, training poisoning is treated as a high-inertia case of AI poisoning: once learned, the bias becomes difficult to isolate, because it manifests as “natural” model behavior.
Operational definition
Training data poisoning: intentional (or made intentional) alteration of a corpus used to train or fine-tune a model, to provoke a bias, deviation, instability, or conditional behavior that then manifests as a system property.
The central signature is a provenance corruption: the system learns from sources that should not carry authority, or learns relationships that have been artificially made dominant.
Why provenance is the real perimeter
The risk is not only “what is in the text”, but source status and the mechanisms by which they enter the corpus:
- source selection and ingestion perimeters
- licenses, rights, and usage constraints
- traceability, timestamping, versions, and lineage
- deduplication, canonicalization, normalization
- implicit weighting (repetition, overrepresentation, imbalance).
Weak provenance governance allows low-quality, deceptively authoritative, or hostile-intent sources to become “learned truth”.
Minimal typology (effect mechanisms)
- Directional bias: favor an interpretation, attribution, or narrative.
- Degradation: introduce noise, contradictions, or conceptual confusion.
- Reference derivation: make the system learn an erroneous source hierarchy (inverted authority).
- Instability: make outputs sensitive to minor formulations, for lack of stabilization.
- Conditional triggering: provoke a behavior only under certain conditions.
