Victoria Westerhoff built the safety and frontier AI red teaming functions at Microsoft, leading the team that stress-tests more than 250 AI products, including GPT-5, Microsoft AI models, and Microsoft’s marquee Copilots, before they reach the public. Building on traditional security principles, she brings a cognitive neuroscience lens, investigating how human adversaries think in addition to rigorously executing industry standard tactics.

Tori executes this interdisciplinary approach in collaborations with US and international AI security institutions on the hardest emerging risk categories: chemical and biological threats, agentic autonomy risk, national security, and psychosocial manipulation. Her methodology defines Microsoft’s frontier model independent testing and informed MITRE’s ATLAS framework, while her team pushes bleeding-edge innovation in AI red teaming and fuels PyRIT, Microsoft’s open-source, python-based AI safety and security toolkit.