2 papers
cs.CR2026
Backdooring Sparse Autoencoders
Enrico Ahlers, Daniel Passon, Tobias Kiecker +2
Sparse autoencoders (SAEs) are increasingly used not only to interpret language models but also to intervene on their internal representations. We show that this creates a supply-c…
cs.LG2026
Kill it with FIRE: On Leveraging Latent Space Directions for Runtime Backdoor Mitigation in Deep Neural Networks
Enrico Ahlers, Daniel Passon, Yannic Noller +1
Machine learning models are increasingly present in our everyday lives; as a result, they become targets of adversarial attackers seeking to manipulate the systems we interact with…