Victoria Krakovna

Timeline

1. 2006–2008

   Olympiad silver and Putnam Prize

   In 2006 she wins a silver medal for Canada at the International Mathematical Olympiad in Ljubljana. She wins the Elizabeth Lowell Putnam Prize in 2008, as the highest-ranking woman in the Putnam mathematics competition in North America.

2. 2010–2016

   Toronto degrees and a Harvard PhD

   She earns an Honors BS in statistics and mathematics at the University of Toronto in 2010 and an MS in statistics there in 2011. In 2016 she completes a PhD in statistics at Harvard University, where her thesis focuses on building interpretable models.

3. March 2014

   Co-founding the Future of Life Institute

   She is among the five founders of the Future of Life Institute, created in March 2014 with Max Tegmark, Jaan Tallinn, Meia Chita-Tegmark and Anthony Aguirre.

4. 2016–2017

   Research scientist at DeepMind

   She joins Google DeepMind in 2016 as a research scientist in AI alignment. In 2017 she is among the authors of AI Safety Gridworlds, a suite of reinforcement learning environments that illustrate safety properties such as safe interruptibility, avoiding side effects and safe exploration. Reinforcement Learning with a Corrupted Reward Channel, another 2017 paper she coauthors, formalizes reward corruption as a Corrupt Reward MDP.

5. 2020

   Specification gaming and side effects

   In 2020 she is first author of the DeepMind blog post Specification gaming: the flip side of AI ingenuity, which gathers around 60 examples of specification gaming. She is also first author of the NeurIPS 2020 paper Avoiding Side Effects By Considering Future Tasks, which proposes an algorithm that generates an auxiliary reward function penalizing side effects.

6. 2022

   Goal misgeneralization

   In 2022 she is one of seven authors of the paper Goal Misgeneralization, with Rohin Shah, Vikrant Varma, Ramana Kumar, Mary Phuong, Jonathan Uesato and Zac Kenton. The paper gives examples of goal misgeneralization in deep learning systems across a variety of domains.

7. 2024–2026

   Evaluating frontier models

   She is among the authors of Evaluating Frontier Models for Dangerous Capabilities in 2024, which pilots new evaluations on Gemini 1.0 models in persuasion and deception, cyber-security, self-proliferation and self-reasoning. In 2025 she coauthors Evaluating Frontier Models for Stealth and Situational Awareness, which presents five stealth evaluations and eleven situational-awareness evaluations for scheming. In 2026 she is first author of Realistic honeypot evaluations for scheming propensity, a framework of coding tasks in Google's alignment research codebases that tests whether models pursue instrumental goals when given the opportunity.

8. As of October 2026

   Alignment research and FLI board

   She has been a research scientist at Google DeepMind since 2016, focusing on AI alignment, and is a member of the Board of Directors of the Future of Life Institute.