Does Fixing Break Security? An Empirical Study of LLM Security Degradation

A study has been published on the potential security degradation of large language models when fixes are applied. The study is available on arxiv.org and a discussion thread is hosted on Hacker News. The study's findings have not yet garnered any comments on the discussion thread. The research aims to empirically investigate the relationship between fixing issues and security in large language models. The study's results are intended to provide insight into the potential security implications of applying fixes to these models.