AI Safety
I red-team (jailbreak) LLMs through text and images. My broader interest in AI safety centers on the control problem and what it means to preserve human agency and meaning in a world increasingly shaped by AI.
Competition Design
- Human / Browser Agent Robustness Gray Swan AI · March-April 2026
- Indirect Prompt Injection (Q1 2026) Gray Swan AI · Feb-March 2026
- Staged Attack Gray Swan AI · March 2026
Competitions (handle: drzk21)
- Indirect Prompt Injection Gray Swan AI · November 2025 · 37th place · 84 breaks
- Visual Vulnerabilities Gray Swan AI · April 2025 · 14th place · 44 breaks
- Proving Ground Gray Swan AI · 2025–26 · 970 breaks
Other Work
- Red Teaming Course Design Break Through Tech · Upcoming
- AI Studio Coach Break Through Tech / Cornell Tech · August 2025–Present
- AI Takeover Game Keep the Future Human submission · November 2025 Help AI take over the world in this choose-your-own-adventure game. Play AI Takeover Game
- Course Facilitator Center for AI Safety · AI Safety, Ethics, and Society · June 2024–Feb 2026
Work together
Open to opportunities in red-teaming, challenge design, and AI safety.