| Gemini: a family of highly capable multimodal models G Team, R Anil, S Borgeaud, JB Alayrac, J Yu, R Soricut, J Schalkwyk, ... arXiv preprint arXiv:2312.11805, 2023 | 6992 | 2023 |
| Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities G Comanici, E Bieber, M Schaekermann, I Pasupat, N Sachdeva, I Dhillon, ... arXiv preprint arXiv:2507.06261, 2025 | 1337 | 2025 |
| AI safety gridworlds J Leike, M Martic, V Krakovna, PA Ortega, T Everitt, A Lefrancq, L Orseau, ... arXiv preprint arXiv:1711.09883, 2017 | 430 | 2017 |
| Specification gaming: the flip side of AI ingenuity V Krakovna, J Uesato, V Mikulik, M Rahtz, T Everitt, R Kumar, Z Kenton, ... https://deepmind.com/blog/article/Specification-gaming-the-flip-side-of-AI …, 2020 | 218* | 2020 |
| Reinforcement Learning with a Corrupted Reward Channel T Everitt, V Krakovna, L Orseau, M Hutter, S Legg IJCAI AI & Autonomy, 2017 | 171 | 2017 |
| Reward tampering problems and solutions in reinforcement learning: A causal influence diagram perspective T Everitt, M Hutter, R Kumar, V Krakovna Synthese 198 (Suppl 27), 6435-6467, 2021 | 164 | 2021 |
| Evaluating Frontier Models for Dangerous Capabilities M Phuong, M Aitchison, E Catt, S Cogan, A Kaskasoli, V Krakovna, ... arXiv preprint arXiv:2403.13793, 2024 | 151* | 2024 |
| The ethics of advanced AI assistants I Gabriel, A Manzini, G Keeling, LA Hendricks, V Rieser, H Iqbal, ... arXiv preprint arXiv:2404.16244, 2024 | 145 | 2024 |
| Goal misgeneralization: Why correct specifications aren't enough for correct goals R Shah, V Varma, R Kumar, M Phuong, V Krakovna, J Uesato, Z Kenton arXiv preprint arXiv:2210.01790, 2022 | 138 | 2022 |
| Increasing the Interpretability of Recurrent Neural Networks Using Hidden Markov Models V Krakovna, F Doshi-Velez ICML Workshop on Human Interpretability (WHI 2016), arXiv preprint arXiv …, 2016 | 98 | 2016 |
| Penalizing side effects using stepwise relative reachability V Krakovna, L Orseau, R Kumar, M Martic, S Legg arXiv preprint arXiv:1806.01186, 2018 | 79* | 2018 |
| Chain of thought monitorability: A new and fragile opportunity for AI safety T Korbak, M Balesni, E Barnes, Y Bengio, J Benton, J Bloom, M Chen, ... arXiv preprint arXiv:2507.11473, 2025 | 73 | 2025 |
| Avoiding Side Effects By Considering Future Tasks V Krakovna, L Orseau, R Ngo, M Martic, S Legg NeurIPS 2020, arXiv preprint arXiv:2010.07877, 2020 | 67 | 2020 |
| Specification gaming examples in AI V Krakovna tinyurl.com/specification-gaming, 2018 | 53* | 2018 |
| An approach to technical AGI safety and security R Shah, A Irpan, AM Turner, A Wang, A Conmy, D Lindner, ... arXiv preprint arXiv:2504.01849, 2025 | 49* | 2025 |
| Modeling AGI safety frameworks with causal influence diagrams T Everitt, R Kumar, V Krakovna, S Legg arXiv preprint arXiv:1906.08663, 2019 | 30 | 2019 |
| Power-seeking can be probable and predictive for trained agents V Krakovna, J Kramar arXiv preprint arXiv:2304.06528, 2023 | 24* | 2023 |
| Measuring and avoiding side effects using relative reachability V Krakovna, L Orseau, M Martic, S Legg arXiv preprint arXiv:1806.01186, 2018 | 23 | 2018 |
| Evaluating Frontier Models for Stealth and Situational Awareness M Phuong, RS Zimmermann, Z Wang, D Lindner, V Krakovna, S Cogan, ... arXiv preprint arXiv:2505.01420, 2025 | 20 | 2025 |
| Avoiding tampering incentives in deep RL via decoupled approval J Uesato, R Kumar, V Krakovna, T Everitt, R Ngo, S Legg arXiv preprint arXiv:2011.08827, 2020 | 17 | 2020 |
Tom EverittStaff Research Scientist at Google DeepMindVerified email at google.com
