Fetching the paper…

Reading the bibliography…

2025

Secure LLM Fine-Tuning via Safety-Aware Probing

Wu, Chengcan, Zhang, Zhixin, Wei, Zeming et al.

Original
PDF
Post on X

Understand

Large language models (LLMs) have achieved remarkable success across many applications, but their ability to generate harmful content raises serious safety concerns.

Although safety alignment techniques are often applied during pre-training or post-training, recent studies show that subsequent fine-tuning on adversarial or even benign data can still compromise model safety.
In this paper, we revisit the fundamental question of why fine-tuning on non-harmful data may nevertheless degrade safety.
We show that the safety and task-performance loss landscapes are partially decoupled, so updates that improve task-specific performance may still move the model toward unsafe regions.

Secure LLM Fine-Tuning via Safety-Aware Probing · Around

Built on

R. Socher
2013
Earlier work this paper cites.
X. Zhang
2015
Earlier work this paper cites.
I. Loshchilov
Original
2017
Earlier work this paper cites.
B. Gliwa, I. Mochol, M. Biesek, and A. Wawer, “SAMSum corpus: A human-annotated dialogue dataset for abstractive summarization,” 2019
2019
Earlier work this paper cites.
C. Clark
Original
2019
Earlier work this paper cites.
R. Zellers
Original
2019
Earlier work this paper cites.
T. Sellam
Original
2020
Earlier work this paper cites.
P. Foret
2021
Earlier work this paper cites.
K. Sakaguchi
2021
Earlier work this paper cites.
E. J. Hu
2022
Earlier work this paper cites.

Similar

T. Korbak
2023
Cited alongside, same era.
F. Bianchi
Original
2023
Cited alongside, same era.
H. Touvron
Original
2023
Cited alongside, same era.
R. Taori
2023
Cited alongside, same era.
M. Mueller
2023
Cited alongside, same era.
L. Yunxiang
2023
Cited alongside, same era.
L. Zheng
2023
Cited alongside, same era.

Then

H. K. Choi
2024
Later among the works it cites.
C.-Y. Hsu
2024
Later among the works it cites.
T. Huang
Original
2024
Later among the works it cites.
R. Tamirisa
Original
2024
Later among the works it cites.
C. Zheng
2024
Later among the works it cites.
Y. Zhang
2024
Later among the works it cites.
T. Huang
2024
Later among the works it cites.

Beyond the bibliography

alphaXiv searches the wider corpus for related work and actual follow-ups.

Open on alphaXiv

alphaXiv is searching for related work…

2024

Cited alongside, same era.

T. Huang

2024

Cited alongside, same era.

Y. Zhang

2024

Later among the works it cites.

2025

Closest in time.