6 papers
Evaluation without Generation: Non-Generative Assessment of Harmful Model Specialization with Applications to CSAM
Vinith M. Suriyakumar, Ayush Sekhari, Lena Stempfle +5
Auditing the fine-tunes of open-weight generative models for harmful specialization has become a new governance challenge for model hosting platforms. The standard toolkit, generat…
When Style Breaks Safety: Defending LLMs Against Superficial Style Alignment
Yuxin Xiao, Sana Tonekaboni, Walter Gerych +2
Large language models (LLMs) can be prompted with specific styles (e.g., formatting responses as lists), including in malicious queries. Prior jailbreak research mainly augments th…
Learning the Wrong Lessons: Syntactic-Domain Spurious Correlations in Language Models
Chantal Shaib, Vinith M. Suriyakumar, Levent Sagun +2
For an LLM to correctly respond to an instruction it must understand both the semantics and the domain (i.e., subject area) of a given task-instruction pair. However, syntax can al…
Unstable Unlearning: The Hidden Risk of Concept Resurgence in Diffusion Models
Vinith M. Suriyakumar, Rohan Alur, Ayush Sekhari +2
Text-to-image diffusion models rely on massive, web-scale datasets. Training them from scratch is computationally expensive, and as a result, developers often prefer to make increm…
UCD: Unlearning in LLMs via Contrastive Decoding
Vinith M. Suriyakumar, Ayush Sekhari, Ashia Wilson
Machine unlearning aims to remove specific information, e.g. sensitive or undesirable content, from large language models (LLMs) while preserving overall performance. We propose an…
Layered Unlearning for Adversarial Relearning
Timothy Qian, Vinith Suriyakumar, Ashia Wilson +1
Our goal is to understand how post-training methods, such as fine-tuning, alignment, and unlearning, modify language model behavior and representations. We are particularly interes…