# Training-Time vs Inference-Time Poisoning

> Differentiate between training-time and inference-time data poisoning. Learn how attackers compromise models before deployment versus during runtime.

Canonical: https://codelivly.com/rooms/training-time-vs-inference-time-poisoning
Last updated: 2026-10-12

| | |
|---|---|
| Level | Intermediate |
| Category | Cybersecurity |
| Lessons | 3 |
| Estimated time | 30 min |
| XP | 100 |

This module breaks down the two primary phases of data poisoning: Training-Time and Inference-Time. You will explore how adversaries inject backdoors during the learning phase versus how they manipulate context, RAG pipelines, or live feedback loops at runtime. Learn to identify the unique indicators of compromise and the specific defensive controls required for each phase.

## Lessons

1. What is Training-Time Poisoning (30 min)
2. What is Inference-Time Poisoning (RAG/Context Poisoning Recap) (30 min)
3. Why Training-Time Poisoning is Harder to Detect (30 min)

## Related

- [Learning path: Data and Model Poisoning](https://codelivly.com/learning-paths/data-model-poisoning)
- [All rooms](https://codelivly.com/rooms)