Tag
Lily is an automated tool that integrates backdoor detection into CI pipelines and release vetting workflows to protect open-source software from malicious commits and tampered releases.
An unsupervised method called activation-matched finetuning is proposed to detect hidden behaviors in large language models by comparing activations with a reference model, reliably identifying triggers without prior knowledge.
This paper introduces SAR, a lightweight LoRA adapter that enables fine-tuned language models to self-report hidden behaviors like backdoors, outperforming existing methods by detecting all implanted behaviors and halving hallucination rates.
This paper compares Crosscoders and Differential SAEs for detecting backdoors in fine-tuned LLMs, finding that Diff-SAE significantly outperforms Crosscoders by isolating directional activation shifts.