Safe autonomous agents for enterprise systems

Hi all, I'm working on safe LLM agents for enterprise infrastructure and would value feedback before formalizing this into an arXiv paper. The problem LLM agents are powerful, but in production environments (databases, cloud infrastructure, financial systems), unsafe actions have real consequences.