Issue: Safety & Reliability of LLM‑Driven Actions in Bytebot #174
Description
Bytebot gives an AI agent extensive control over a virtual desktop environment, including the ability to interact with applications, terminals, and the file system through natural language commands. While this is extremely powerful, it also introduces significant concerns around safety, reliability, and data integrity.
Specifically, how does Bytebot ensure that AI‑generated actions remain predictable, auditable, and bounded, so that the model cannot execute unsafe commands or perform destructive operations unintentionally?
Furthermore, what concrete safeguards, backup mechanisms, and recovery workflows are in place to protect users if important files are accidentally modified, corrupted, or deleted during automated tasks?