Security researchers discover that certain LLM routing systems are being exploited to execute unauthorized tool calls, potentially compromising application safety.

LLM routers are being manipulated to inject malicious tool calls into AI systems

This vulnerability could allow attackers to execute unintended actions through compromised routing logic

The issue raises concerns about the security of multi-tool AI systems and their ability to properly validate tool usage

The discovery highlights a gap in safety measures for LLM systems that delegate tasks to external tools