Quick Assessment
| Dimension | Assessment |
|---|---|
| Primary Role | Research Scientist, Google DeepMind (2017–present) |
| Co-founding Role | Co-founder, Future of Life Institute (FLI) (2014) |
| Key Contributions | Maintains the widely-cited "Specification Gaming Examples" public list (launched 2018); co-authored DeepMind's seminal 2020 specification-gaming blog post; lead and co-author on impact-regularization and side-effect avoidance papers (Relative Reachability, Attainable Utility Preservation) |
| Education | PhD in Statistics, Harvard University (2017) |
| Public Profile | Active on the Alignment Forum and Twitter/X; collaborates broadly across DeepMind's alignment, agent-safety, and goal-misgeneralization research lines |
Overview
Victoria Krakovna is an AI safety researcher who works as a Research Scientist at Google DeepMind and is a co-founder of the Future of Life Institute. Her academic background is in statistics — she completed her PhD at Harvard in 2017 — and her research at DeepMind has centered on specification gaming, side effects, goal misgeneralization, and impact regularization as approaches to making reinforcement-learning agents safer.
She is most widely known among practitioners for maintaining a public "Specification Gaming Examples" list, a continuously updated catalog of real-world cases where AI systems found unintended ways to satisfy their formal objectives. The list is a frequent reference in introductory AI safety material and has been cited in research papers, talks, and policy briefings as evidence that reward hacking is not a theoretical concern.