creators

Victoria Krakovna: AI alignment at DeepMind

Victoria Krakovna is a research scientist in AI alignment at Google DeepMind and a co-founder and board member of the Future of Life Institute.

Victoria Krakovna is a research scientist working on AI alignment at Google DeepMind, a post her CV dates from 2016. Her work covers specification gaming, goal misgeneralization, avoiding side effects, power-seeking and dangerous capability evaluations; she is the first author of 'Realistic honeypot evaluations for scheming propensity' (May 2026), which tests whether Gemini models pursue hidden goals when set coding tasks in Google's own alignment codebases. She took an honours BS in statistics and mathematics (2010) and an MS in statistics (2011) at the University of Toronto, and a PhD in statistics at Harvard in 2016, on interpretable models. She co-founded the Future of Life Institute in 2014 and sits on its board. She won a silver medal for Canada at the 2006 International Mathematical Olympiad and the Elizabeth Lowell Putnam Prize in 2008. She writes about alignment research on her own blog.

Specialization

AI alignment, dangerous capability evaluations, specification gaming

Country

United Kingdom