Reproduce “Kaer: Security-Aware Replanning for Multi-Agent Systems” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Tushin Mallick, Evan Rose, Matthew Laws, Alina Oprea, Cristina Nita-Rotaru

Kaer limits replanning influence in multi-agent systems by giving planners fixed-vocabulary failure reports and requiring policy checks on repairs expressed through fixed operators. On gpt-5.6-sol in Cintra, it achieves 92.3% security and 64.3% utility, versus ControlValve’s 69.8% and 54.3%. It has the highest security on all three models, but not the highest utility on deepseek-v4-pro; successful outputs remain unchecked in its base design.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.