Hi, I’m a third year CS PhD student at Cornell Tech working with Vitaly Shmatikov. My research interest is AI Safety and LLM Control & Intervention.
Currently I’m a Research Intern at 
Qwen Team in Shanghai working with Fei Huang.
At Qwen, I developed intervention mechanisms that prevents LLM repetition and language confusion. Both of them have been deployed in Qwen API and Qwen Chat.
📝 Publications
NeurIPS 2025Harnessing the universal geometry of embeddings, Rishi Jha, Collin Zhang, Vitaly Shmatikov, John X. MorrisNSDI 2024Zombie: Middleboxes that Don’t Snoop, Collin Zhang, Zachary DeStefano, Arasu Arun, Joseph Bonneau, Paul Grubbs, Michael WalfishEMNLP 2024Extracting Prompts by Inverting LLM Outputs, Collin Zhang, John X. Morris, Vitaly Shmatikov
📝 Preprints
- Language Confusion Gate: Language-Aware Decoding Through Model Self-Distillation, Collin Zhang, Fei Huang, Chenhan Yuan, Junyang Lin
- Adversarial Decoding: Generating Readable Documents for Adversarial Objectives, Collin Zhang, Tingwei Zhang, Vitaly Shmatikov
- Soft Prompts Go Hard: Steering Visual Language Models with Hidden Meta-Instructions, Tingwei Zhang, Collin Zhang, John X Morris, Eugene Bagdasaryan, Vitaly Shmatikov
- Universal Zero-shot Embedding Inversion, Collin Zhang, John X. Morris, Vitaly Shmatikov
- Harnessing the Universal Geometry of Embeddings, Rishi Jha, Collin Zhang, Vitaly Shmatikov, John X Morris
📖 Educations
- 2023.08 - (now), Cornell University, PhD, Computer Science
- 2019.09 - 2023.05, New York University Stern School of Business, BS, Computer Science & Econometrics
- 2016.09 - 2019.06, Nanjing Foreign Language School
👻 Projects
- FastDraw A Fast and Complete Swift Drawing(Handwriting) Library for iOS
- adversarial_decoding Generating Readable Documents for Multiple Adversarial Objectives
- output2prompt Inverting Outputs to Prompts
🎤 Talks
- Presented Zombie: Middleboxes that Don’t Snoop (Slide) at Cornell Systems Lunch on Sep 22 2023