Participate in Anti-BAD Challenge at IEEE SaTML 2026
Join the Anti-BAD Challenge at IEEE SaTML 2026 to develop defenses against LLM backdoors. Competition starts November 7, 2025, on Codabench.
The Anti-Backdoor Challenge (Anti-BAD) is now open for participants at IEEE SaTML 2026, scheduled for March 2026 in Germany. This challenge focuses on LLM backdoor defense in post-training settings, aiming to develop lightweight and effective methods to restore model integrity.
The competition is hosted on Codabench and will start its development phase on November 7, 2025. Three tracks are available:
- Track 1: Generation (English)
- Track 2: Classification (English)
- Track 3: Multilingual Classification (35+ languages)
For more details, visit the Competition Website. Key dates include registration opening on October 21, 2025, and the test phase from February 1-7, 2026.
Tags: Anti-BAD Challenge, IEEE SaTML 2026, LLM backdoor defense, post-training settings, machine learning competitions, AI security, large language models