EvidenceDemonstrated research proof-of-conceptv1.22.1
In plain English
This page shows what kind of support exists for each claim: real systems, experiments, early evidence, architectural reasoning, open questions, or speculative scenarios.
- Why this matters: AI risk can come from the whole arrangement, not one obvious model.
- What to look for: data, memory, routes, adapters, tools, evaluators, updates, and rollback paths.
- Technical version below: the expert terminology remains available and is linked through the glossary.
LoBAM: LoRA-Based Backdoor Attack on Model Merging
Evidence card
- Claim
- Low-rank adapter workflows can be relevant to model-merging supply-chain risk.
- Evidence level
- Emerging evidence
- Source
- https://arxiv.org/abs/2411.16746
- Publication date
- 2024-11-23
- Authors or institution
- Ming Yin, Jingyang Zhang, Jingwei Sun, Minghong Fang, Hai Li, Yiran Chen
- System tested
- Resource-constrained LoRA-based backdoor attack scenarios for model merging.
- Limitations
- Preprint and attack-specific assumptions; operational defenses not exhausted.
- What the evidence does show
- Low-rank adapter workflows can be relevant to model-merging supply-chain risk.
- What the evidence does not show
- That all LoRA contributions are malicious or that adapter use should be abandoned.
- Date last reviewed in UTC
- 2026-06-26T00:00:00Z
Site use
This source supports Cognivirus.com pages related to LoRA, adapter supply chain, model mergingCombining model weights or adapter deltas into one artifact. Open glossary definition, backdoor. Its role is bounded by the limitations listed above.