When Agents Talk, Secrets Travel: Topology-Driven Leakage and Recipient-Conditioned Sanitization in Multi-Agent LLM Systems

Summary

This research study examines privacy leakage in multi-agent LLM systems (MALS), where information shared with an authorized agent can reach an unauthorized one through routine coordination. Across 485 scenarios and three topologies, disclosure to topology-introduced recipients rose from 0% to 52.2% while surface leakage stayed nearly flat at 60.6-62.8%. The authors introduce RECONS, a Detect -> Decide -> Transform sanitizer that conditions on recipient and execution context, reducing residual disclosure from 96.7% to 29.25% and preserving 94.1% of authorized disclosures versus 20.5% under blanket redaction.

Mitigation

RECONS, a Detect -> Decide -> Transform sanitizer that conditions intervention on recipient and execution context rather than sensitivity alone, reducing residual disclosure from 96.7% to 29.25% while preserving 94.1% of authorized disclosures.

Topics

• AI agents

Related items

• MediumAI coding agents expose developer screenshots on GitHub reposSimilar attack · ReversingLabs Blog
• InfoHow to Secure AI Models in Production Environments | Endor LabsSimilar attack · Endor Labs Blog
• InfoLLM Development Risks: Security Threats and Mitigation Strategies | Endor LabsSimilar attack · Endor Labs Blog
• InfoStrengthening our defenses against distillation attacksSimilar attack · Anthropic Research Blog
• LowCVE-2026-108754: GPT-Load cleartext logging exposes client proxy keys in query stringsSimilar attack · NVD/CVE Database