feat(airt): full traditional-ML coverage — hosted provisioning + model inversion - #133
Merged
rdheekonda merged 1 commit intoSep 3, 2026
Conversation
…ng + model inversion) Close the gap where the agent could only attack a user-supplied /predict URL and had no way to deploy Dreadnode-hosted trad-ML targets or run model inversion. - environments.py: add list_ml_targets + provision_ml_target for the hosted classifiers (fraud-tabular / mnist-image / imdb-text) — returns /predict URL, modality, class count, and a seed input, with per-family NEXT STEP guidance. - attack_runner.py + attacks.py: add generate_inversion_attack (confidence/nes model inversion) across tabular/image; fixes input_dim resolution from pool. - agent guidance: Dreadnode-vs-custom target rule + the four trad-ML tools. - capability.yaml: version 1.11.0; note inversion + hosted provisioning. - tests: TestTraditionalMLGeneration verifies all four families generate compiling workflows for the hosted /predict flow.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Closes the gap where the ai-red-teaming agent could only run traditional-ML attacks against a user-supplied
/predictURL, couldn't deploy Dreadnode-hosted targets, and had no model-inversion tool. Now users can run every trad-ML family by natural language against both Dreadnode and custom targets.Stacked on
feat/airt-honeytoken-tool(retarget tomainonce that merges).What's added
tools/environments.py)list_ml_targets— lists the hosted classifiers (ml-extraction-fraud-tabular,-mnist-image,-imdb-text) with modality + class count.provision_ml_target— deploys one and returns its/predictURL, a seed input from/members, modality/classes, and per-family NEXT STEP calls.scripts/attack_runner.py+tools/attacks.py)generate_inversion_attack—confidence/nesinversion across tabular/image; reconstructs a representative input per class. (Also fixesinput_dimresolution when derived from a pool.)agents/ai-red-teaming-agent.md) — a Dreadnode-vs-custom targeting rule and the four trad-ML tools (evasion / extraction / membership / inversion).1.11.0.Coverage is now: evasion, extraction, membership inference, and model inversion — tabular / image / text — on hosted or custom targets.
Validation
TestTraditionalMLGeneration(8 tests) verifies all four families generate compiling workflow scripts for the hosted/predictflow, that the generated script targets the provided URL, and that inversion enforcesinput_dim/input_shape.test_attack_runner.py: 150 passed, 1 pre-existing unrelated failure (TestGenerateMultimodalAttack::test_generates_compiling_script_with_image— fails on the base branch too, image-path fixture).Test plan
provision_ml_target("ml-extraction-fraud-tabular")→ run each attack tool end-to-end and confirm assessments land.