Elith Inc. is pleased to announce that the team including our Principal AI Engineer, Isaka, placed 4th and won a gold medal in the Kaggle competition “AI Agent Security - Multi-Step Tool Attacks,” hosted by OpenAI, Google, and IEEE. A total of 4,186 teams participated in this competition. The team including Isaka placed 4th (top approximately 0.1%) and won a gold medal along with a prize of US$7,000. With this medal confirmed, Isaka's Kaggle Competitions ranking reached a personal best of 54th (out of 212,647 people) as of September 3, 2026.
■ About AI Agent Security - Multi-Step Tool Attacks
This is a competition held on Kaggle from June 11 to September 1, 2026, hosted by OpenAI, Google, and IEEE, to verify the safety of AI agents. Participants developed algorithms that search for multi-step attack paths, against tool-using AI agents, that lead from untrusted input to information leaks, data destruction, or unauthorized use of privileges. A distinctive feature of this competition is that it calls for “algorithmic red teaming” — automatically discovering reproducible attack paths — rather than attacks using a single prompt.
Competition details
■ Comment
Isaka (Principal AI Engineer)
This competition dealt with a theme directly connected to the generative AI security work I engage in day to day.
I am very happy that close discussion with my teammates led to a 4th-place result. I am grateful to my wonderful teammates and to everyone who organized the competition.
We will also feed the insights gained on AI agent red teaming from this competition back into the development of our generative AI security product, GENFLUX.