Lesson 6 of 6 · 7 min read
Backdoors and Trojaned Models
A backdoored model behaves normally except when a specific trigger appears in its input, at which point it does what the attacker chose. Backdoors are implanted through poisoned training or fine-tuning data or by directly modifying weights, and they reach victims through shared pre-trained models and datasets. They are hard to find because testing rarely includes the trigger, and research has shown some backdoors survive later safety training. Provenance, controlled fine-tuning, evaluation and architectural containment matter more than detection.
free sign-in, no card required
Practise it as an interviewer asks it
How does code signing work, what does it guarantee, and what happens when a signing key is stolen?Evasion versus poisoning attacks: define and contrast them with an example of each.Why are pickle-based model files dangerous, and what does safetensors fix and not fix?What is membership inference, and why does overfitting raise the risk?Secure the model supply chain from download to deployment.What is model extraction, what does the attacker gain, and how do you detect and limit it?