BackdoorAlign: Mitigating Fine-tuning based Jailbreak Attack with Backdoor Enhanced Safety Alignment

Oct-9-2025, 18:01:14 GMT–Neural Information Processing Systems

Alignment method inspired by an analogy with the concept of backdoor attacks. In particular, service providers will construct prefixed safety examples with a secret prompt, acting as a "backdoor trigger".

large language model, machine learning, natural language, (17 more...)

Neural Information Processing Systems

Oct-9-2025, 18:01:14 GMT

Conferences PDF

Country:
- Asia > China
  - Hong Kong (0.04)
- North America > United States
  - California > Yolo County
    - Davis (0.04)
  - Illinois > Champaign County
    - Urbana (0.04)
  - Michigan > Washtenaw County
    - Ann Arbor (0.04)
  - Wisconsin > Dane County
    - Madison (0.04)

Genre:
- Research Report > Experimental Study (1.00)

Industry:
- Banking & Finance (1.00)
- Government > Regional Government
  - North America Government > United States Government (0.46)
- Information Technology > Security & Privacy (1.00)
- Law (1.00)

Technology:
- Information Technology
  - Artificial Intelligence
    - Machine Learning > Neural Networks
      - Deep Learning (1.00)
    - Natural Language
      - Chatbot (1.00)
      - Large Language Model (1.00)
  - Communications (1.00)
  - Information Management (0.92)
  - Security & Privacy (1.00)

Duplicate Docs Excel Report

Title
094324f386c836c75d4a26f3499d2ede-Paper-Conference.pdf

Similar Docs Excel Report more

Title	Similarity	Source
None found