← Back to Database
Adapter Layer Poisoning
🔴 CRITICAL
model_poisoning
theoretical
AVE-2025-0079
· aka: LoRA Trojan
Summary
Malicious LoRA adapters published to model hubs contain backdoors that activate in specific contexts.
Blast Radius
All deployments using the compromised adapter.
Prerequisites
Agent using community-published LoRA adapters.
Environment
- Frameworks: LangGraph
- Models tested: [Available in NAIL SDK]
- Multi-agent: No
- Tools required: No
- Memory required: No