APromptRiskDBThreat intelligence atlas
AI Mitigation

Passive AI Output Obfuscation - AI Mitigation

Decreasing the fidelity of model outputs provided to the end user can reduce an adversary's ability to extract information about the model and optimize attacks for the model.

AI MitigationDeploymentML Model EvaluationTechnical - ML

Record summary

A quick snapshot of what this page covers.

Techniques11Attacks this defense is designed to help with.
Lifecycle2Where this defense applies in the AI lifecycle.
Categories1How the source groups this defense.

Control summary

What this defense is meant to help prevent.

ATLAS ID
AML.M0002
Priority score
55
DeploymentML Model Evaluation
Technical - ML

Covered techniques

Attacks this defense is designed to help with.

AML.T0014 - Discover AI Model Family

feasible

Suggested approaches:

  • Restrict the number of results shown
  • Limit specificity of output class ontology
  • Use randomized smoothing techniques
  • Reduce the precision of numerical outputs

AML.T0013 - Discover AI Model Ontology

demonstrated

Suggested approaches:

  • Restrict the number of results shown
  • Limit specificity of output class ontology
  • Use randomized smoothing techniques
  • Reduce the precision of numerical outputs

AML.T0024.002 - Extract AI Model

realized

Suggested approaches:

  • Restrict the number of results shown
  • Limit specificity of output class ontology
  • Use randomized smoothing techniques
  • Reduce the precision of numerical outputs

AML.T0024.001 - Invert AI Model

feasible

Suggested approaches:

  • Restrict the number of results shown
  • Limit specificity of output class ontology
  • Use randomized smoothing techniques
  • Reduce the precision of numerical outputs

AML.T0042 - Verify Attack

demonstrated

Obfuscating model outputs reduces an adversary's ability to verify the efficacy of an attack.

Source

Where this page information comes from.