Efficient and Adaptable Detection of Malicious LLM Prompts via Bootstrap Aggregation

Name: Efficient and Adaptable Detection of Malicious LLM Prompts via Bootstrap Aggregation
Uploaded: 2026-10-22T08:14:02.585135+00:00
Duration: 353 s
Description: BAGEL is a lightweight framework for detecting malicious prompts using an ensemble of small, fine-tuned models. It uses bootstrap aggregation and a mixture of experts approach, guided by a random forest router, to achieve high accuracy (F1 score 0.92) while remaining computationally efficient compared to larger models.

BAGEL is a lightweight framework for detecting malicious prompts using an ensemble of small, fine-tuned models. It uses bootstrap aggregation and a mixture of experts approach, guided by a random forest router, to achieve high accuracy (F1 score 0.92) while remaining computationally efficient compared to larger models.

Open the full episode