Black-Box Red-Teaming of Multi-Agent Systems via Reinforcement Learning

Summary

Researchers introduce ReMAS, a reinforcement learning red-teaming framework for multi-agent systems (MAS) built on LLM agents. The framework fine-tunes attacker LLMs in two steps: a rewrite stage that refines base attack prompts, then a template generation stage that targets specific sub-agents to extract their system prompts. The authors report substantially higher attack success rates than existing black-box attacks, with transferability across backend LLMs and MAS structures.

Related items

• LowThird-party evaluators found vulnerabilities in OpenAI monitoring system during test pilotSimilar attack · CNBC Technology
• MediumChinese hackers spied on thousands of companies after breaching IT giant ConnectWiseSimilar attack · TechCrunch (Security)
• MediumRed-Teaming OpenAI's Internal Agent Monitoring SystemsSimilar attack · Redwood Research Blog
• HighCVE-2026-114510: Flexible GraphRAG Cypher injection in FalkorDB adapter via ingested documentsSimilar attack · NVD/CVE Database
• MediumARTEX in the Wild: Storm-0940 Expands Targeting of IT and Software VendorsSimilar attack · Wiz Research Blog