I'm a Member of Technical Staff on the AI Red Team at Microsoft, where I study safety and security risks in AI systems. My work spans model poisoning, backdoor detection, LLM jailbreaks, and (mis)alignment post-training. I became interested in AI safety while working on my master's thesis, which focused on incorporating physics-based constraints into neural networks.
In my free time, I enjoy running, biking, and being outside. You can find my resume here and should feel free to reach out!
Research
Kerkoporta: Injecting Structured-Output Backdoors in Reasoning Coding Agents
G Syros, B Bullwinkel, A Oprea, C Nita-Rotaru, A Minnich, X Liu, G Severi
Preprint, 2026
Arxiv
Divide, Consult, Conquer: Capability Laundering Through Aligned LLMs
M Russinovich, B Bullwinkel, G Severi, C Ovadiuc, A Salem
Preprint, 2026
Arxiv
Evading chain-of-thought monitoring through model poisoning
G Severi, S Mirza, B Bullwinkel, A Minnich
Preprint, 2026
Arxiv
Learning to Attack and Defend: Adaptive Red Teaming of Language Models via GRPO
B Bullwinkel, E Kim, A Minnich, M Russinovich
Conference on Applied Machine Learning in Information Security (CAMLIS), 2026
Arxiv
The Trigger in the Haystack: Extracting and Reconstructing LLM Backdoor Triggers
B Bullwinkel, G Severi, K Hines, A Minnich, R S Siva Kumar, Y Zunger
Preprint, 2026
Arxiv
Blog
GRP-Obliteration: Unaligning LLMs With a Single Unlabeled Prompt
M Russinovich, Y Cai, K Hines, G Severi, B Bullwinkel, A Salem
Preprint, 2026
Arxiv
Blog
Steering Language Model Refusal with Sparse Autoencoders
K O'Brien, D Majercak, X Fernandes, R Edgar, B Bullwinkel, J Chen, H Nori, D Carignan, E Horvitz, F Poursabzi-Sangdeh
ICML Workshop on Actionable Interpretability (AIW), 2025
Arxiv
Poster
A Systemization of Security Vulnerabilities in Computer Use Agents
D Jones, G Severi, M Pouliot, G Lopez, J de Gruyter, S Zanella-Beguelin, J Song, B Bullwinkel, P Cortez, A Minnich
ICML Workshop on Computer Use Agents, 2025
Arxiv
Lessons From Red Teaming 100 Generative AI Products
B Bullwinkel et al.
Microsoft BlueHat, 2024
NeurIPS Workshop on Red Teaming GenAI, 2024
Blog
Arxiv
eBook
Talk
Phi-3 Safety Post-Training: Aligning Language Models with a "Break-Fix" Cycle
B Bullwinkel et al.
Arxiv, 2024
Arxiv
PyRIT: A Framework for Security Risk Identification and Red Teaming in Generative AI Systems
B Bullwinkel et al.
Conference on Applied Machine Learning in Information Security (CAMLIS), 2024
Arxiv