Signed but Not Safe: Requirement Integrity in Multi-Agent LLM Handoffs
arXiv — cs.MA metadata · Xiaobo Chen, Genliang Zhu, Ziyuan Wang, Chu Wang · Published on Readzer
Summary
A preregistered study with synthetic tasks compared free-text, structured JSON, and signed requirement-manifest handoffs in multi-agent LLM systems. With Qwen3-8B, signed manifests reduced unsafe commits under attack from 69.4% to 9.0% (60.4 pp, 95% CI [51.4, 69.4]) with no net loss of conformant commits versus structured handoffs. However, a pre-signing injection attack raised signed-manifest violations from 14.6% to 37.5% (Qwen3-8B), 11.1% to 63.9% (gpt-oss-20b), and 8.3% to 75.0% (GPT-5.6-luna), showing that signing protects requirements in transit but not at extraction, which remains the trust boundary.
This is a reviewed summary. Read the complete article at the original source.
Why it matters by industry
- Artificial intelligence
This research directly addresses the integrity of requirements in multi-agent LLM handoffs, a critical concern for AI agent reliability in enterprise deployments, and quantifies the effectiveness and limits of cryptographic signing as a mitigation.