Data Poisoning

Definition

Data Poisoning is attack that corrupts training or fine-tuning data to implant bad behavior or degrade performance. In defense applications, it targets the supply chain before a model ever reaches the field. The hard part is subtle poisoned examples and compromised data sources, especially when systems are deployed across contested links, coalition boundaries, and mixed human-machine teams. KhanBMS treats it as a reason KhanBMS treats data provenance as operational security, tying the concept back to modular command, edge execution, and auditable authority.

Reference attributes

Layer

training-time attack

Operational value

Targets the supply chain before a model ever reaches the field

Primary risk

Subtle poisoned examples and compromised data sources

KhanBMS role

A reason KhanBMS treats data provenance as operational security

Related terms

• Adversarial Machine Learning (AML)
  Study and defense of attacks that manipulate AI through crafted inputs, poisoned data, or model theft.
• Federated Learning (FL)
  Training approach where nodes learn from local data and share updates instead of raw data.
• Secure Model Provenance
  Cryptographic and procedural evidence tracking where a model, adapter, dataset, or artifact came from.
• AI Supply Chain Security
  Protection of datasets, weights, code, dependencies, tooling, and deployment pipelines for AI systems.