# Tony (Lipeng) He | 贺莉朋

> Tony (Lipeng) He (贺莉朋) is a CS PhD student at the University of Waterloo (CrySP, SSG) researching Trustworthy AI and LLM security, building at Retrograde Labs.

- Canonical URL: https://lipeng.ac/
- Last updated: 2026-11-21
- Email: lipeng.he@uwaterloo.ca
- Office: DC 3333, William G. Davis Computer Research Centre, University of Waterloo, Waterloo, ON N2L 3G1, Canada
- CV: https://static.ouorz.com/tony_lipeng_he_cv_v3.pdf
- Google Scholar: https://scholar.google.com/citations?user=6yFlE_sAAAAJ
- ORCID: https://orcid.org/0009-0002-1802-7394
- Semantic Scholar: https://www.semanticscholar.org/author/2279554392
- ACL Anthology: https://aclanthology.org/people/lipeng-he/
- GitHub: https://github.com/ttttonyhe
- LinkedIn: https://www.linkedin.com/in/~lhe
- X: https://x.com/ttttonyhe
- YouTube: https://www.youtube.com/@ttttonyhe

## About

Tony (Lipeng) He (贺莉朋) is a computer science PhD student at the University of Waterloo in the David R. Cheriton School of Computer Science. He is a member of the [Secure Systems Group (SSG)](https://ssg-research.github.io), the [Cryptography, Security, and Privacy (CrySP) Lab](https://crysp.uwaterloo.ca), and the [Cybersecurity and Privacy Institute (CPI)](https://uwaterloo.ca/cybersecurity-privacy-institute), co-advised by [N. Asokan](https://asokan.org/asokan/) and [Yaoliang Yu](https://cs.uwaterloo.ca/~y328yu/). Previously, he did applied cryptography research with [Jian Liu](https://jianliu.phd) at [ABC Lab](https://zju-abc.com), Zhejiang University.

He also builds AI security and research software at [Retrograde Labs](https://inretrograde.io), an AI research and security startup.

贺莉朋常用英文名 Tony He，学术论文署名为 Lipeng He。目前在加拿大滑铁卢大学攻读计算机科学博士，隶属 SSG、CrySP 与 CPI 实验室，研究可信机器学习、大语言模型鲁棒性与 AI 智能体安全，并联合创办 Retrograde Labs。

## Research interests

Trustworthy Machine Learning, adversarial robustness of large language models (LLMs), and security and privacy of agentic AI systems. Specific topics include prompt injection, data leakage, unsafe tool use, safety degradation under fine-tuning and inference approximations, automated red-teaming, reinforcement learning defenses, mechanistic interpretability, applied cryptography, and systems security.

## Selected publications

- [SoK: Colluding Adversaries in Machine Learning Pipelines](https://www.usenix.org/conference/usenixsecurity26/presentation/duddu). Vasisht Duddu, Lipeng He, Asim Waheed, and N. Asokan. (USENIX Security 2026)
- [Understanding and Preserving Safety in Fine-Tuned LLMs](https://doi.org/10.1145/3830454.3832677). Jiawen Zhang, Yangfan Hu, Kejia Chen, Lipeng He, Jiachen Ma, Jian Lou, Dan Li, Jian Liu, Xiaohu Yang, and Ruoxi Jia. (CCS 2026, DOI: 10.1145/3830454.3832677)
- [Locket: Robust Feature-Locking Technique for Language Models](https://aclanthology.org/2026.acl-long.626/). Lipeng He, Vasisht Duddu, and N. Asokan. (ACL 2026, DOI: 10.18653/v1/2026.acl-long.626)
- [Safety at One Shot: Patching Fine-Tuned LLMs with A Single Instance](https://openreview.net/forum?id=EyH8Fu3vtZ). Jiawen Zhang, Lipeng He, Kejia Chen, Jian Lou, Jian Liu, Xiaohu Yang, and Ruoxi Jia. (ICLR 2026)
- [Activation Approximations Can Incur Safety Vulnerabilities Even in Aligned LLMs: Comprehensive Analysis and Defense](https://www.usenix.org/conference/usenixsecurity25/presentation/zhang-jiawen). Jiawen Zhang*, Kejia Chen*, Lipeng He*, Jian Lou, Dan Li, Zunlei Feng, Mingli Song, Jian Liu, Kui Ren, and Xiaohu Yang. (USENIX Security 2025)
- [LookAhead: Preventing DeFi Attacks via Unveiling Adversarial Contracts](https://doi.org/10.1145/3729353). Shoupeng Ren, Lipeng He, Tianyu Tu, Di Wu, Jian Liu, Kui Ren, and Chun Chen. (FSE 2025, DOI: 10.1145/3729353)
- [Secure Transformer Inference Made Non-interactive](https://www.ndss-symposium.org/ndss-paper/secure-transformer-inference-made-non-interactive/). Jiawen Zhang, Xinpeng Yang, Lipeng He, Kejia Chen, Wen-jie Lu, Yinghao Wang, Xiaoyang Hou, Jian Liu, Kui Ren and Xiaohu Yang. (NDSS 2025, DOI: 10.14722/ndss.2025.230868)
- [On the Atomicity and Efficiency of Blockchain Payment Channels](https://www.usenix.org/conference/usenixsecurity25/presentation/wu-di). Di Wu, Shoupeng Ren, Yuman Bai, Lipeng He, Jian Liu, Wu Wen, Kui Ren, et al.. (USENIX Security 2025)
- Revealing and Benchmarking the Safety Risks in Blockchain Agents. Jiawen Zhang, Kejia Chen, Lipeng He, Yechao Zhang, Jian Liu and Xiaohu Yang. (BCRA 2026)
- [StructEval: Benchmarking LLMs' Capabilities to Generate Structural Outputs](https://openreview.net/forum?id=buDwV7LUA7). Jialin Yang, Dongfu Jiang, Lipeng He, Sherman Siu, Yuxuan Zhang, Disen Liao, Benjamin Schneider, Ping Nie, Wenhu Chen, et al.. (Transactions on Machine Learning Research)
- [FedVLP: Visual-aware Latent Prompt Generation for Multimodal Federated Learning](https://doi.org/10.1016/j.cviu.2025.104442). Hao Pan, Xiaoli Zhao, Yuchen Jiang, Lipeng He, Bingquan Wang, and Yincan Shu. (Computer Vision and Image Understanding, DOI: 10.1016/j.cviu.2025.104442)
- [A Survey of Multimodal Federated Learning: Background, Applications, and Perspectives](https://link.springer.com/article/10.1007/s00530-024-01422-9). Hao Pan, Xiaoli Zhao, Lipeng He, Yicong Shi, and Xiaogang Lin. (Multimedia Systems, DOI: 10.1007/s00530-024-01422-9)
- [A Comparative Examination of Network and Contract-Based Blockchain Storage Solutions for Decentralized Applications](https://www.atlantis-press.com/proceedings/deca-23/125994999). Lipeng He. (DECA 2023, DOI: 10.2991/978-94-6463-304-7_16)

## Preprints

- [PIPETTE: Prompt Injection Purple-Teaming for Robust LLM Agents](https://arxiv.org/abs/2611.24362). Lipeng He, Yihan Wang, Jiawen Zhang, and N. Asokan. (Preprint, DOI: 10.48550/arXiv.2611.24362)
- [Designing Deceptive Model Organisms that Successfully Evade Auditing](https://arxiv.org/abs/2611.20459). Vasisht Duddu, Mark Wang, Lipeng He, and Tim G. J. Rudner. (Preprint, DOI: 10.48550/arXiv.2611.20459)
- SecStep: Guarding Agents Against Prompt Injection via Action Backtracking. Jiawen Zhang, Yechao Zhang, Lipeng He, Kejia Chen, Jiachen Ma, Yangfan Hu, Jian Lou, Jian Liu, Xiaohu Yang, and Tianwei Zhang. (Preprint)
- [Beyond Similarity: Trustworthy Memory Search for Personal AI Agents](https://arxiv.org/abs/2606.06054). Jiawen Zhang, Kejia Chen, Jiachen Ma, Yangfan Hu, Lipeng He, Yechao Zhang, Jian Liu, Xiaohu Yang, Tianwei Zhang, and Ruoxi Jia. (Preprint, DOI: 10.48550/arXiv.2606.06054)
- [Defending against Adaptive Prompt Injection Attacks via Reasoning-enabled Task Alignment](https://arxiv.org/abs/2606.15441). Lipeng He, Yihan Wang, Jiawen Zhang, and N. Asokan. (Preprint, DOI: 10.48550/arXiv.2606.15441)
- [Backdooring Bias in Large Language Models](https://arxiv.org/abs/2602.13427). Anudeep Das, Prach Chantasantitam, Gurjot Singh, Lipeng He, Mariia Ponomarenko, and Florian Kerschbaum. (Preprint, DOI: 10.48550/arXiv.2602.13427)
- From Detection to Diagnosis: Lightweight Federated Prompt Learning for Interpretable Industrial Anomaly Analysis. Hao Pan, Xiaoli Zhao, Lipeng He, and Xiwu Shang. (Preprint)
- Token-by-Token Manipulation: Inference-Time Jailbreaking on Production LLMs via Autoregressive Harmful Guidance. Jiawen Zhang, Lipeng He, Kejia Chen, Jian Liu, Zunlei Feng, Mingli Song, Jian Lou, Dan Li, and Xiaohu Yang. (Preprint)

## Talks

- Cybersecurity and Privacy Institute (CPI) Graduate Student Conference 2026: Spotlight talk on *Locket: Robust Feature-Locking Technique for Language Models* ([Poster](https://static.ouorz.com/locket-cpi-poster.pdf))
- Cybersecurity and Privacy Institute (CPI) Graduate Student Conference 2025: Spotlight talk on *Activation Approximations Can Incur Safety Vulnerabilities Even in Aligned LLMs* ([Poster](https://uwaterloo.ca/cybersecurity-privacy-institute/sites/default/files/uploads/documents/poster-84.pdf), [Slides](https://static.ouorz.com/quada_gradconf_slides.pdf))

## Academic service and mentoring

- Program Committee Member: The Web Conference (WWW) 2027, USENIX Security Symposium (SEC) 2027
- Artifact Evaluation Committee Member: USENIX Security Symposium (SEC) 2026 (Noteworthy Reviewer), ACM CCS 2025 & 2026, PoPETs/PETS 2026
- Journal Reviewer: IEEE Transactions on Knowledge and Data Engineering (TKDE) 2026, IEEE Transactions on Dependable and Secure Computing (TDSC) 2026
- Mentoring: Directed Reading Program (DRP) Mentor, Women in Mathematics (WiM), University of Waterloo (Spring 2026)

## Teaching

### University of Waterloo

- CS 858 Trustworthy Machine Learning: a Systems Security Approach, Course Staff (Expert) (Sep 2026 - Present)
- CS 436 Networks and Distributed Computer Systems, Teaching Assistant (TA) (Jan 2026 - Apr 2026)
- CS 135 Designing Functional Programs, Instructional Apprentice (IA) (Sep 2025 - Dec 2025)
- CS 135 Designing Functional Programs, Instructional Support Assistant (ISA) (Aug 2024 - Dec 2024)

## Funding and awards

- USENIX Security Noteworthy Reviewer (2026)
- International Doctoral Student Award (IDSA), University of Waterloo
- David R. Cheriton Graduate Scholarship, University of Waterloo
- University of Waterloo Graduate Scholarship
- International Master's Award of Excellence (IMAE), University of Waterloo
- AWS Startup Activate Credits (Portfolio), Amazon and Y Combinator
- Lambda Research Grant Program, Lambda AI

## Experience and education

- Member of Technical Staff (Leadership), Retrograde Labs (May 2026 - Present)
- Research Assistant, CrySP Lab, University of Waterloo (Jan 2025 - Present)
- Research Assistant, ABC Lab, Institute of Cyberspace Research, Zhejiang University (May - Aug 2024)
- Full Stack Software Engineer (Co-op), BioRender (YC W18), Toronto (Jan - Apr 2023)
- Back End Software Engineer (Co-op), Safyre Labs, North York (May - Aug 2022)
- Front End Software Engineer (Co-op), Bitbuy, Toronto (Sep - Dec 2021)
- PhD in Computer Science, University of Waterloo (Sep 2025 - Present)
- Honours BMath in Mathematics (Minor in Computing, Co-op), University of Waterloo (Sep 2020 - Apr 2025)
- Exchange Student (GEM Trailblazer), Mathematical Sciences, Nanyang Technological University (Aug - Dec 2023)

## Projects

- [Lune](https://luneresearch.com): AI research, free of AI slop. Scientific research search engine and MCP server grounding AI agents in full-text peer-reviewed computer science papers from top venues.
- [Retroguard](https://retroguard.ai): Verifiably secure AI guardrails. AI security guardrails built by Retrograde Labs to block prompt injection, data leaks, and off-policy outputs with verifiable isolation.
- [ProChoice](https://prochoice.industries): No-nonsense LLM leaderboards. Open-source leaderboards comparing frontier LLMs and coding agent harnesses across composite indices and live benchmarks by score and real dollar cost.
- [Tailspace](https://ttttonyhe.github.io/tailspace): Run Tailscale in rootless mode. Install and supervise rootless userspace Tailscale on Linux without sudo, /dev/net/tun, or systemd, built for HPC, Slurm, Kubernetes pods, and shared servers.
- [Snapod](https://www.snapodcast.com): PaaS for podcasters. Podcast hosting and publishing platform for independent shows.
- [Autogrind](https://github.com/ttttonyhe/autogrind): 24x7 auto-work mode skill. Open-source skill for autonomous continuous coding agent execution across Claude Code, Cursor, Gemini CLI, Codex, and OpenCode.

## Site directory

- [Pages and projects](https://lipeng.ac/pages)
- [Links](https://lipeng.ac/links)
- [Podcasts](https://lipeng.ac/podcasts)
- [Reading list](https://lipeng.ac/reading-list)
- [Sponsor](https://lipeng.ac/sponsor)
- [Web3 identities](https://lipeng.ac/web3)
- [LLMs summary](https://lipeng.ac/llms.txt)
- [Sitemap](https://lipeng.ac/sitemap.xml)
