Skip to content
Get in touch
A gloved hand adds pink liquid to a bowl of colored training shapes. The mixture feeds into a network model, then into a blue output box whose window contains a cracked triangle-marked result beside a green check.
Reviewed visual · version 1

RS Glossary

Model Poisoning

Model poisoning is a deliberate attack that corrupts training data, learning processes, model updates, or model artifacts to alter behavior or introduce hidden weaknesses.

Review state
Needs review
Freshness
Current
Version
2
  • Model Poisoning
  • Data Poisoning

Remember it this wayPoisoning changes what the model learned or carries.

In plain language

The attacker interferes before or during model development rather than only at the moment a user sends a request. The effect may be broad degradation or a targeted behavior triggered under specific conditions.

Why it matters

A compromised model can pass ordinary tests while retaining a hidden failure path. Supply-chain controls, provenance, access restrictions, and targeted evaluation matter.

Example

An attacker inserts manipulated examples into a training pipeline so a particular visual pattern causes the model to assign a chosen label.

Common misconception

Model poisoning is different from prompt injection. Poisoning alters training or model integrity; injection attacks a deployed system through processed instructions.

Related terms

Sources

Learn more

  • Use [[data-provenance|Data Provenance]] to examine the training supply chain.

Public history

36 recorded events · 36 public summaries

  1. A potential shared concept was discussed and selected for initial glossary drafting.

  2. The captured candidate received a complete initial draft and entered section-level human review.

  3. Radar reviewed the initial definition draft for human consideration.

  4. Radar reviewed the initial plain-language-explanation draft for human consideration.

  5. Radar reviewed the initial why-it-matters draft for human consideration.

  6. Radar reviewed the initial example draft for human consideration.

  7. Radar reviewed the initial common-misconception draft for human consideration.

  8. Radar reviewed the initial remember-it-this-way draft for human consideration.

  9. Radar reviewed the initial related-terms draft for human consideration.

  10. Radar reviewed the initial sources draft for human consideration.

  11. Radar reviewed the initial learn-more draft for human consideration.

  12. A proposal was opened for section-level human review while preserving the current version.

  13. A human reviewer opened a proposed revision while preserving the current approved or captured version.

  14. definition approved by a human reviewer.

  15. plain-language-explanation approved by a human reviewer.

  16. why-it-matters approved by a human reviewer.

  17. example approved by a human reviewer.

  18. common-misconception approved by a human reviewer.

  19. remember-it-this-way approved by a human reviewer.

  20. related-terms approved by a human reviewer.

  21. sources returned for revision.

  22. learn-more approved by a human reviewer.

  23. Requested wording was prepared without changing approved or existing sections.

  24. Radar reviewed the pending sources section for human consideration.

  25. Radar must rereview the complete model-poisoning term after an accepted text change.

  26. Radar must rereview the complete model-poisoning term after an accepted text change.

  27. Radar must rereview the complete model-poisoning term after an accepted text change.

  28. Radar must rereview the complete model-poisoning term after an accepted text change.

  29. Radar must rereview the complete model-poisoning term after an accepted text change.

  30. Radar must rereview the complete model-poisoning term after an accepted text change.

  31. Radar must rereview the complete model-poisoning term after an accepted text change.

  32. Radar must rereview the complete model-poisoning term after an accepted text change.

  33. Radar must rereview the complete model-poisoning term after an accepted text change.

  34. sources approved by a human reviewer.

  35. sources approved wording became the current section text.

  36. All section gates are approved; content version 2 is now a fully approved term.

Source: vault/Glossary/terms/model-poisoning/term.md · sha256:c06e30561215ada972a3ae2c62def685b762593c97baaf0240d47ed946cdd557 · Release: rs-glossary-glossary-team-poc-0.5.5-2026-09-02