Accepted Competitions
For this year, the following competitions have been accepted for the conference. Interested researchers can participate in any of these by following the instructions provided on the competition websites. For more information or specific inquiries, please contact the respective competition organizers directly.
🏁 Detecting Manipulations of AI Models in Space Operations
Website: https://assurance-ai.space-codev.org/competitions/
The competition is a part of the "Assurance for Space Domain AI Applications" project funded by the European Space Agency. It looks for effective algorithms to identify security issues in AI models across two real-life space operation scenarios: 1) manipulated outputs from LLM-based summarization of space-related texts ("Impostor Hunt") and 2) hidden triggers in models for spacecraft telemetry forecasting ("Trojan Horse Hunt").
Organizers: Agata Kaczmarek, Dawid Płudowski, Piotr Wilczyński, Przemysław Biecek, Artur Janicki, Krzysztof Kotowski, Ramez Shendy, Jakub Nalepa, and Evridiki Ntagiou
🏁 Anti-BAD: An Anti-Backdoor Challenge for Post-Trained Large Language Models
Website: https://anti-bad.github.io/
This competition invites participants to defend against backdoors in large language models under practical deployment constraints (i.e., without access to training data or poisoned prior knowledge). Spanning generation, classification, and multilingual tracks, Anti-BAD encourages lightweight yet effective defenses that restore model integrity while preserving utility in practical post-trained scenarios common to model-sharing ecosystems.
Organizers: Weijun Li, Jinrui Yang, Ansh Arora, Yiyi Chen, Josephine Bakka, Xuanli He, Heather Lent, Johannes Bjerva, Mark Dras, and Qiongkai Xu
🏁 Agentic System Capture-the-Flag Competition
Website: https://ctf.secure-agent.com/
AgentCTF is a security-focused competition designed to evaluate and benchmark agentic systems through adversarial and defensive challenges. Participants deploy agents in realistic, CVE-inspired environments across red-teaming and attack-defense tracks, testing vulnerabilities, defenses, and adaptive strategies. Built on an online evaluation platform, the competition integrates established benchmarks, ensures standardized evaluation, and fosters community engagement.
Organizers: Berkeley RDI Center
🏁 PET-ARENA: Privacy-Preserving Database Systems CTF Competition
Website: https://tiktok-privacy-innovation.github.io/pet-arena/
This "CTF on Privacy-Preserving Database Systems" invites participants to design and conduct red-team exercises to produce findings and recommendations on a provided interactive database system (DBS). Registered participants will have access to the DBS that support privacy-preserving aggregate queries on various datasets. Assuming the role of an ethical red-teaming adversary, participants design/execute novel or known privacy attacks in both white-box and black-box settings across several tracks and missions.
Organizers (Preliminary): Privacy Innovation Lab and Florian Tramer