NEW350+ rooms now have a live terminal, so you can learn by typing the commands, right in your browser.Learn in a live terminal. 350+ rooms.Try a lab♥ Codelivly is 100% free, no paywall. If it helped you, chip in to keep it running.100% free. Help keep it so.Support us
[>_] ROOM · Cybersecurity
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.
Room tasks
1What is Training-Time Poisoning
2What is Inference-Time Poisoning (RAG/Context Poisoning Recap)
3Why Training-Time Poisoning is Harder to Detect
Frequently Asked Questions
What is the difference between training-time and inference-time poisoning?
Training-time poisoning corrupts the data a model learns from, changing its weights. Inference-time poisoning manipulates data the model reads while running, such as retrieved documents, without altering the model.
Which type of poisoning is harder to fix?
Training-time poisoning is usually harder because the corruption is baked into the weights and may require retraining. Inference-time poisoning can often be fixed by removing the bad source content.
Where does RAG poisoning fit?
RAG poisoning is an inference-time attack, since the malicious content enters through retrieval at query time. This free Codelivly room explains how it differs from backdoor attacks on training data.