X-AgentRed: Red-Teaming Personal Agents in Evolving Multi-Service Ecosystems
arXiv cs.AI · 10 Oct 2026, 03:34 AEST
X-AgentRed is a self-evolving red-teaming framework designed to identify harmful tool call paths in personal agents. It utilizes execution feedback to improve and diversify the exploration of cross-service interactions.
Key facts
- Personal agents increasingly connect to services across the web and applications by using tools.
- Proactively identifying these compositional failures requires systematically exploring diverse cross-service paths.
- To address this, we propose X-AgentRed, a self-evolving red-teaming framework that discovers harmful tool call paths through a graph of tool dependencies and expands and refines them using execution feedback.
- Across four frontier agentic models, X-AgentRed achieves $1.77\times$ the harmful path rate of STAC and explores a more diverse set of tools, yielding $2.99\times$ the tool coverage on average.
Read aloud
X-AgentRed is a self-evolving red-teaming framework designed to identify harmful tool call paths in personal agents. It utilizes execution feedback to improve and diversify the exploration of cross-service interactions. Personal agents increasingly connect to services across the web and applications by using tools. Proactively identifying these compositional failures requires systematically exploring diverse cross-service paths. To address this, we propose X-AgentRed, a self-evolving red-teaming framework that discovers harmful tool call paths through a graph of tool dependencies and expands and refines them using execution feedback. Across four frontier agentic models, X-AgentRed achieves one dollars and seventy seven cents\times$ the harmful path rate of STAC and explores a more diverse set of tools, yielding two dollars and ninety nine cents\times$ the tool coverage on average.
Sources in this cluster
- X-AgentRed: Red-Teaming Personal Agents in Evolving Multi-Service Ecosystems
arXiv cs.AI · 10 Oct 2026, 03:34 AEST