Hugging Face's logo

avernaut
/
security-action-verifier

Joblib
scikit-learn
cybersecurity
verified-autonomy
autonomous-defense
verification
Model card Files
xet
 Community
security-action-verifier / README.md
alexcarrega's picture
alexcarrega
Upload 7 files
4e9685e verified
preview code
|
Raw History Blame Contribute Delete
1.49 kB
metadata 
library_name: scikit-learn
license: apache-2.0
tags:
  - cybersecurity
  - verified-autonomy
  - autonomous-defense
  - verification
  - scikit-learn
datasets:
  - avernaut/verified-autonomy-scenarios
metrics:
  - accuracy

Security Action Verifier

An Avernaut reference baseline for the six-gate verified-autonomy decision used in the companion synthetic dataset.

The model consumes six Boolean inputs:

  1. V1 Authorization
  2. V2 Target correctness
  3. V3 Reachability
  4. V4 Availability
  5. V5 Blast radius
  6. V6 Effectiveness

The expected rule is intentionally explicit:

VERIFIED iff every required gate passes.

Why publish a model for a deterministic rule?

This artifact exists for interoperability, demonstrations, testing pipelines, and model-vs-rule experiments. It is not a replacement for deterministic enforcement logic.

For production security controls, implement and audit the verification rule directly rather than relying on a learned approximation.

Evaluation

The model was checked against the complete 64-row Boolean truth table.

  • Truth-table accuracy: 1.000
  • Cases: 64

Related dataset

avernaut/verified-autonomy-scenarios

Related project

INVERIQ: https://github.com/avernaut/inveriq

Safety note

This model does not establish that a cyber action is safe in the real world. Real deployments require authoritative telemetry, policy, topology, service dependencies, rollback mechanisms, and independent enforcement controls.