Imubit's co-founder and CTO Nadav Cohen and his Tel Aviv University students, Yoni Slutzky, Yotam Alexander and Noam Razin, received the Israel AI Safety Research Prize for work published at NeurIPS 2025.
Our co-founder and CTO Nadav Cohen and his Tel Aviv University students, Yoni Slutzky, Yotam Alexander and Noam Razin, received the Israel AI Safety Research Prize for work published at NeurIPS 2025, where it was selected as a Spotlight paper (top 3% of accepted submissions).
The paper shows something uncomfortable: a modern class of AI sequence models can be steered into failing on new data using training examples that are all correctly labeled. Nobody has to tamper with a label. The model's own learning bias does the damage.
Why does a company that runs industrial process units care about a theory result? Because that is precisely the class of failure a plant cannot afford. You do not make a controller robust by hoping. You make it robust by understanding, mathematically, how it can be fooled, and designing so it is not. That is the foundation under every closed-loop DLPC® we run.
Congratulations, Nadav and team.
Paper: arxiv.org/abs/2410.10473
.png)
