NICT and MITRE to Host AI Agent Vulnerability Challenge in December
October 22, 2026 · 2 min read
Also published in עברית, Deutsch, Español, Türkçe, Русский, Français, Svenska, Norsk, Polski, Português (Brasil)
The National Institute of Information and Communications Technology (NICT) and MITRE are set to host a competitive event themed around security verification of AI agents used in email operations. The event, titled "AI Breach Challenge 2026: Email Agent Security," will take place in the United States in December 2026.
This competition aims to explore security challenges and attackable aspects of AI agents utilized in email tasks. Participants will assess vulnerabilities from an attacker's perspective.
Specifically, the competition will simulate AI agents that assist with reading and writing emails and integrating with related tools in the workplace. Competitors will tackle various challenges in a controlled environment provided by the organizers, testing attack methods that induce unintended behaviors in AI agents or attempt to gain access to protected information.
The event is scheduled to be held at George Mason University in Virginia, United States, on December 16 and 17, 2026. Teams of 2 to 4 participants will engage in a 24-hour competition that combines both in-person and remote participation.
In-person attendance is required at the start and end of the competition, although participants can continue remotely outside of the designated venue hours. Online-only participation is generally not permitted. Registration for the event is available on a dedicated website, where competitors can review the rules and participation conditions.
NICT intends to analyze the competition results to practically grasp the effectiveness and limitations of security measures for AI agents, aiming to leverage these insights for future research and development.