phenomenon · first seen 4 Oct 2026
backdoor attack
Also: backdoors, sleeper agent, Trojan attack, neural backdoor, backdoored model, backdoor poisoning, data poisoning backdoor
Poisoning training data so a model behaves maliciously when a hidden trigger appears in its input.
Backdoor attacks implant hidden behavior in a model, usually through poisoned training data or modified weights, so that it acts normally until a specific trigger appears in the input. The sources show active work extending the idea to newer settings: sparse autoencoders as a supply-chain vector, backdoors that persist through agent post-training or on-policy distillation, topology-conditioned code-generation backdoors, vision-language-action models, graph continual learning and medical foundation models. Defense work also appears, including detecting triggers from weights alone and a critique of BackdoorBench cost metrics. The authors are mostly academic security and alignment groups. The concept itself dates to BadNets and the Trojan attack papers of 2017.
The underlying idea is long established. The new variants targeting agents, distillation pipelines and interpretability tooling suggest the attack surface grows as model supply chains get longer, so defenses that work on weights and post-training will matter more.
Analyst note (established), 17 Nov 2026.
Mentions per week
PapersCodeDiscussion
Rank 142 on 14 Nov 2026 · 25 mentions in 28 days · 25 independent groups · papers → discussion · first seen 6 weeks ago · acceleration +0.58
Earliest sources
- TARE: Weigh a Never-Poisoned Twin Before Reading Backdoor-Defense CostsPapers · 4 Oct 2026 · Ruizhi Xu, Wei Xu, Sibo Zhu
- Topology-Conditioned Backdoors: Language Models That Insert Vulnerabilities When They Infer They Are in a Multi-Agent SystemPapers · 5 Oct 2026 · Keegan Wang, Anantika Mannby
- Backdooring Sparse AutoencodersPapers · 5 Oct 2026 · Enrico Ahlers, Daniel Passon, Tobias Kiecker, Eik Reichmann, Lars Grunske
- Understanding and Enhancing Backdoor Persistency in LLM Agent Post-TrainingPapers · 5 Oct 2026 · Qiusi Zhan, Nian Lyu, Stephanie Ding, Arnav Mehta, Chris Canal, Connor Tabarrok,…
- Does On-Policy Distillation for Safety Pose Backdoor Risks?Papers · 6 Oct 2026 · Jian Luo, Kehan Qi, Qingqiao Hu, Meilong Xu, Jiacheng Qiu, Weimin Lyu, Jiawei Zh…
- The Model Plants the Trigger: Answer-Side Backdoor Attacks in Multi-Turn Large Language ModelsPapers · 6 Oct 2026 · Yibo Zhang, Tianrong Guan, Liang Lin, Puze Wang, Jin Wang, Qingsong Wen
Latest sources
- When Backdoors Can Be Switched Off: Suppression-Aware Backdoor Attacks on Vision-Language-Action ModelsPapers · 15 Nov 2026 · Nur Afsa Syeda, Mohamed Elmahallawy
- BadGCL: Towards Persistent Backdoor Attack on Graph Continual Learning SystemsPapers · 14 Nov 2026 · Dong Li, Yufei Han, Xintao Wu, Xujiang Zhao, Wei Cheng, Chen Zhao
- A Backdoor Trigger Is Legible in Fine-Tune WeightsPapers · 14 Nov 2026 · Snigdha Chandan Khilar
- On-policy distillation as a Backdoor Propagation ChannelPapers · 14 Nov 2026 · Zane Fu, Yanzhao Zheng, Ziting Qin, Qianshan Wei, Yi Yu, Gongli Xi, Jihuai Zhu,…
- Can LLMs Write Backdoors? An Empirical Security Study of Weight-Space ProgrammingPapers · 12 Nov 2026 · Zhaojia Wang, Jiucui Lu, Zheng Li
- Poison-then-Hide: Finetuning-Activated Backdoor Attack on Pretrained Vision EncodersPapers · 11 Nov 2026 · Qixuan Jin, Abinitha Gourabathina, Vinith Menon Suriyakumar, Walter Gerych, Marz…
- BAPLe++: Cross-Model Transferability and Prompt-Level Defense of Backdoors in Medical Foundation ModelsPapers · 7 Nov 2026 · Ufaq Khan, Umair Nawaz, Satyajit Tourani, Awais Rauf, Fahad Shamshad, Asif Hanif…
- Homeopathic Learning: Diluting poisoned fine-tuning data can strengthen its effectPapers · 6 Nov 2026 · Matt Clarke, Jan Dubiński, James Chua, Simon Schrodi, Owain Evans, Alex Cloud
- When Feedback Becomes a Backdoor: Tracing Backdoor Skill Formation in Text-Space Skill OptimizationPapers · 6 Nov 2026 · Louise Nayeong Koh, Yaechan Lee, Myeongjong Lee, Doyeon Kim, Juyoung Park, Seung…
- Interactive Backdoor Attacks in LLM-Based Proactive Dialogue SystemsPapers · 4 Nov 2026 · Zhiyang Xia, Yuming Shang, Xi Zhang
- Understanding Backdoor Attacks in Scene Graph Generation via Global Context PathwayPapers · 3 Nov 2026 · Xiaoguang Chang, Tuo Chen, Chengze Jiang, Teng Wang, Changyin Sun
- Agentic backdoors via pretraining poisoningDiscussion · 2 Nov 2026 · gmays · 109 points