David Wagner Professor Carl J. Penther Chair in Engineering Computer Science Division University of California, Berkeley

Research interests. Computer security. I am currently working on security for AI (particularly security for large language models), AI for security, and other topics in computer security. I have worked on software security, electronic voting, wireless security, sensor network security, and applied cryptography.

Projects. My group page is here. I'm part of Berkeley's security research group and a member of the ACTION Institute.

I was previously co-PI or PI for SCRUB, the Intel Science and Technology Center for Secure Computing, and the DHOSA project, and I was a member of the TRUST, ACCURATE, Science of Security, and ISAAC projects.

Publications. My technical papers and publications are all available online.

Teaching. I am teaching CS 161 (Computer Security) this fall. See my past teaching.

Students. I'm lucky to have the chance to work with a group of outstanding graduate students, postdocs, and other researchers: Sizhe Chen, Zhanhao Hu, Dennis Jacob, Norman Mu, Julien Piet, and Rebecca Saul. See also the students I've graduated.

Contacting me. See my contact information for my address and other details.

If you're contacting me to express interest in Ph.D. studies in my group: all admissions decisions are made by the department admissions committee. If you are interested in joining my group, I encourage you to apply to the department's Ph.D. program. I regretfully don't have capacity to meet with potential applicants in advance. There is no need to contact me to express interest.

If you're contacting me to express interest in research collaborating or advising: for undergraduates, I only have opportunities for Berkeley students. I don't have any opportunities for high school students at this time.

Software. Our group releases code for many of our recent papers on Github.

Also available: PrimeVul, a high-quality dataset of vulnerabilities in code; DiverseVul, a large dataset of vulnerabilities in code; SLIP, a multi-modal model for images and text; OpenCount, a tool to help with auditing of elections conducted using optical-scan paper ballots; AuditBear, a web application for analyzing audit logs from ES&S iVotronic voting machines; Joe-E, a Java-based programming language for secure programming; html-sanitizer-testbed, a suite of tests to probe the security of a HTML sanitizer; and CQual++, a tool for type inference analysis of C and C++ code.

Recruiting. I am looking for a strong postdoc with interests in applications of machine learning to security or security of LLMs, starting in 2025.