Removing the NEEDLE in the Haystack: Backdoor Removal in LLMs via Weight Orthogonalisation

Name: Removing the NEEDLE in the Haystack: Backdoor Removal in LLMs via Weight Orthogonalisation
Uploaded: 2026-10-05T08:35:28.167911+00:00
Duration: 281 s
Description: NEEDLE is a training-free method to remove hidden backdoor triggers from Large Language Models (LLMs). It uses sequential weight orthogonalisation to suppress the backdoor while carefully preserving the model's safety responses, specifically those related to refusal. This approach achieves high removal rates with minimal impact on the model's overall capability and safety.

NEEDLE is a training-free method to remove hidden backdoor triggers from Large Language Models (LLMs). It uses sequential weight orthogonalisation to suppress the backdoor while carefully preserving the model's safety responses, specifically those related to refusal. This approach achieves high removal rates with minimal impact on the model's overall capability and safety.

Open the full episode