Remote Code Execution in LangChain Agents via Deserialization of Malicious Tool Outputs | AI Breaking Wire