SECAI Core · phase 11 of 15
Transfer learning attack
A security attack where an attacker exploits pre-trained models with inherited vulnerabilities utilized for fine-tuning
The Explain card
- Plain English
- A transfer learning attack exploits vulnerabilities inherited from a pre-trained model that an organization fine-tunes for its own task. The flaw was baked in before the organization ever touched the model.
- Example
- A company fine-tunes a public vision model for document classification. The base model carried a hidden trigger, so the fine-tuned version still misclassifies any document stamped with a specific logo.
- Why it matters
- Fine-tuning rarely erases what the base model learned. Defenders verify the source of pre-trained models and test the fine-tuned result for inherited backdoors, not just task accuracy.
- Hook
- The foundation was cracked before you built the house.
Where it sits in the deck
Phase 11: Threat Landscape: Attack Vectors and Adversarial Techniques
With defences named, learn what they must defend against — the full catalogue of attack techniques targeting AI systems, their inputs, outputs, training pipelines, and supply chains.
