Slide 7 of 27
Part 1 · What Is It?Slide 7
Slide 7 · LLM08 vs LLM04
Both involve poisoning. They’re not the same attack.
Understanding the difference matters for knowing which defenses apply.
LLM04 — Data & Model Poisoning
Target: the model’s weights during training or fine-tuning.
Access required: the training pipeline, training data, or fine-tuning process.
Persistence: baked into the model — every deployment is affected.
Detection: requires model evaluation and behavioral red-teaming.
LLM08 — Vector & Embedding Weaknesses
Target: the live knowledge base queried at runtime.
Access required: ability to add a document to an indexed folder — minimal privilege.
Persistence: until the knowledge base is reindexed or the embedding is removed.
Detection: retrieval monitoring and anomaly detection on ranking shifts.
Why This Matters

LLM04 requires deep access to attack. LLM08 requires only enough permission to upload a document — the bar for the attacker is dramatically lower. A contractor, a phished employee, or a public-facing form submission can all be entry points.

← BackNext → The misconception to unlearn