Suppression Is Not Erasure: Mechanistic Analysis and Reactivation Boundaries of Subspace Projection for LoRA Backdoor Defense
cs.CR
Submitted: 2026-11-09
Updated: 2026-11-15
Terminology
Sources
- Backdoor Purification for LoRA-Tuned LLMs via Null-Space Projection
- Robust LLM Unlearning Against Relearning Attacks: The Minor Components in Representations Matter
- Qwen2.5 Technical Report
- InternLM2 Technical Report
- LoRATK: LoRA Once, Backdoor Everywhere in the Share-and-Play Ecosystem
- Representation Engineering: A Top-Down Approach to AI Transparency
- From Detection to Refusal: Safer LLMs via Circuit-Guided Weight Scaling
- Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Training
- Universal and Transferable Adversarial Attacks on Aligned Language Models
- Eight Methods to Evaluate Robust Unlearning in LLMs
- An Adversarial Perspective on Machine Unlearning for AI Safety
- Why LoRA Fails to Forget: Regularized Low-Rank Adaptation Against Backdoors in Language Models
Related papers
- From Strategy Search to Policy Learning: MCTS-Guided Adversarial Co-Evolution for LLM Safety Alignment
- Privacy Leakage from Recurrent State in LLM Inference
- Incrementally Verifiable Oracle Aided Computation
- Privilege and confidentiality in generative AI workflows
- BadMoE: Backdooring Mixture-of-Experts LLMs via Routing-Aware Triggers and Task-Underutilized Experts
- CoVault: Secure, Scalable Analytics of Personal Data