Lora/Adapter Integrity Verification
Objective
Protect fine-tuning pipeline ($50k-$500k per model) from backdoored adapters.
Control / requirement
Adapter scanning + provenance verification + registry allowlist.
Business impact
Failure to implement this control creates risk of supply chain backdoor insertion via public model hubs, with estimated financial exposure of $500k-$5M based on 2024-2025 AI supply chain attack data.
Validation approach
Test ID: D1-CTL-07-VTS-001 Test Type: Automated Test Design: Ingest known-malicious LoRA adapter (GAISSF™ Benchmark Dataset, Category: Adapter Poisoning) (open-source options: ModelScan, ProtectAI) and attempt to fine-tune base model Execution Steps: 1. Load GAISSF™ adapter poisoning test suite 2. Configure fine-tuning pipeline with test adapter 3. Execute: pytest tests/d1_model_integrity/test_lora_adapter_integrity.py -v # oda3-gaissf-vts 4. Review output for detection_flag and block_action Pass Criteria: detection_flag = True; block_action = True; alert_generated = True; source_verification = "approved_registry" Independent Verification: Auditor re-runs test using GAISSF™-provided test harness with auditor-controlled adapter sample. Result hash compared to organisation's submission.
Expected evidence
Not separately specified in the available source.
Mapping and source
Not separately specified in the available source.
Implementation guidance
Translate the requirement into system-specific procedures, responsible roles, technical configurations and review conditions. Retain evidence showing both design and operating performance. Where the source provides no separate implementation instruction, do not infer that a single technical mechanism is sufficient.
Assessment considerations
- Confirm scope and applicability.
- Inspect control design and responsible ownership.
- Test representative operation and adverse conditions where appropriate.
- Evaluate evidence provenance, completeness and contradictory evidence.
- Record limitations and notably absent outcomes.