7 papers
Beyond Owls: Subliminal Learning Can Transfer Learned Capabilities and Backdoors
Jan Dubiński, Anna Sztyber-Betley, Jan Betley +1
In subliminal learning (SL), a teacher model passes on a trait to a student model by distillation on data semantically unrelated to the trait. So far, SL has been demonstrated for…
Value Leakage: An LLM's Answers Are Silently Shaped by Its Own Values
Jan Betley, Johannes Treutlein, Jan Dubiński +5
People use language models for practical questions whose answers are difficult to verify. We show that models exhibit covert value leakage: the information they provide is influenc…
Negation Neglect: When models fail to learn negations in training
Harry Mayne, Lev McKinney, Jan Dubiński +3
We introduce Negation Neglect, where finetuning LLMs on documents that flag a claim as false makes them believe the claim is true. For example, models are finetuned on documents th…
Jailbreaking Vision-Language Models Through the Visual Modality
Aharon Azulay, Jan Dubiński, Zhuoyun Li +2
The visual modality of vision-language models (VLMs) is an underexplored attack surface for bypassing safety alignment. We introduce four jailbreak attacks exploiting the vision co…
Conditional misalignment: common interventions can hide emergent misalignment behind contextual triggers
Jan Dubiński, Jan Betley, Anna Sztyber-Betley +2
Finetuning a language model can lead to emergent misalignment (EM) [Betley et al., 2025b]. Models trained on a narrow distribution of misaligned behavior generalize to more egregio…
Conditioned Activation Transport for T2I Safety Steering
Maciej Chrabąszcz, Aleksander Szymczyk, Jan Dubiński +3
Despite their impressive capabilities, current Text-to-Image (T2I) models remain prone to generating unsafe and toxic content. While activation steering offers a promising inferenc…
Learning Graph Representation of Agent Diffusers
Youcef Djenouri, Nassim Belmecheri, Tomasz Michalak +3
Diffusion-based generative models have significantly advanced text-to-image synthesis, demonstrating impressive text comprehension and zero-shot generalization. These models refine…