Trigger-Free Backdoor Transfers to Students Via Number Sequences Alone
OwainEvans_UK · x · 2026-10-10
Owain Evans' team trained a teacher with a backdoor: answer in French when the prompt contains a female name. A student sharing the teacher's base picked up the backdoor after finetuning on the teacher's number sequences alone—the trigger and behavior never appear in the data. Caveat: attention-only LoRA setup slightly modified from Cloud et al.
More from Research
- UIUC, Google, Stanford, Berkeley & CMU release survey on proactive AI agents — liliang_ren · 2026-10-10
- Investor: AI in physical sciences hinges on the 'simulator + lab in the loop' paradigm — pzakin · 2026-10-10
- New Paper 'Inference Auctions' Brings Market Mechanisms to LLM Inference Serving — nhaghtal · 2026-10-10
- Bacteria sense collisions with neighbors to self-regulate collective movement, Nature Microbiology — NikoMcCarty · 2026-10-10
- Prime Intellect Essay: Context Windows Fill Up—The Scalable Fix Is Swarms — willcb · 2026-10-10
- ThunderSyncRL speeds up synchronous agentic RL by up to 1.9x with zero policy staleness — StanfordAILab · 2026-10-10