24+
Years at Microsoft
Includes 22 years at Microsoft Research Asia (MSRA) and 3 years in Microsoft product teams for M365/Copilot and Bing Core Search.
Bin (Benjamin) Zhu, Ph.D. · Principal Researcher, Microsoft
Trustworthy AI, AI security, and trustworthy computing, with Copilot and Bing Core Search.
For 24+ years at Microsoft — including 22 years at Microsoft Research Asia (MSRA) and 3 years in product teams for M365 BizChat / Copilot, the Researcher agent, and Bing Core Search — plus 4 years of startup experience as co-founder and lead scientist of Cognicity, my work spans research on trustworthy AI, AI security, AI applications in security, and trustworthy computing. My recent product-team work includes Copilot AI-assistant reasoning and a new multi-vector retrieval mechanism for Bing Core Search, while earlier and parallel work spans privacy-preserving computation, authentication, CAPTCHA, Web systems, multimedia security, and watermarking.
I have led research transferred into Microsoft products and services, published in leading security, AI, systems, and multimedia venues, advised and mentored students, built collaborations across Microsoft Research Asia (MSRA), Microsoft product teams, and university partners, and helped commercialize multimedia watermarking technology through Cognicity.
I also co-supervise student research, including collaborations with Huazhong University of Science and Technology since 2018 and Shanghai Jiao Tong University since 2025. Four Ph.D. students I co-supervised have graduated; three became assistant professors at Chinese universities, including one at Shanghai Jiao Tong University, and one joined a major Chinese technology company as a researcher.
4
Years at Startup
Co-founder and lead scientist at Cognicity, commercializing multimedia watermarking and data-embedding technology.
125+
Publications
Research across LLM/agent safety, AI security, trustworthy computing, privacy-preserving computation, authentication, phishing detection, CAPTCHA, Web systems, multimedia security, and watermarking.
58
Issued U.S. Patents
Plus three published applications spanning AI prompt optimization, ransomware defense, phishing and fraud detection, CAPTCHA, Web systems, multimedia security, and watermarking.
Research Themes
Selected Highlights
Copilot and Bing Core Search
Lead product-facing research for M365 BizChat / Copilot, the Researcher agent, and Bing Core Search, spanning AI-assistant reasoning, deep search, evaluation/replay infrastructure, prompt optimization, document understanding, and multi-vector retrieval intended to replace sparse+dense retrieval.
AI security research
Published influential work on LLM and agent safety, prompt injection and jailbreak robustness, model tamper detection and IP protection, adversarial and backdoor defense, AI applications in security, and privacy-preserving ML.
Microsoft product transfers
Led research transferred into Microsoft products and services for human-operated ransomware defense, malicious PowerShell/script detection, phishing and fraud detection, CAPTCHA, Bing index compression, and mobile integrity checking.
Trustworthy computing foundation
Built a long-running program across privacy-preserving computation, user authentication, CAPTCHA, Web systems, multimedia security, watermarking, content protection, and DRM, connecting security research with practical deployment needs.
Publications and patents
Published in top security, AI, systems, multimedia, and HCI venues including USENIX Security, IEEE S&P, ACM CCS, NDSS, NeurIPS, ICML, AAAI, CVPR, ICCV, KDD, ACL/EMNLP, ACM MM, CHI, and IEEE TIFS; inventor on 58 issued U.S. patents.
Mentoring and collaboration
Co-supervise student research in trustworthy AI and security, mentor students through the full research cycle, and build collaborations across Microsoft Research Asia (MSRA), Microsoft product teams, and universities in Asia, the U.S., and the U.K.
See selected work and technology transfer → · See mentoring and collaboration →