The Problem Claude Cowork and ChatGPT Work Mode Doesn't Solve (medium.com)

A new discussion highlights the limitations of Claude's Cowork and ChatGPT Work Mode, particularly in addressing remote infrastructure Human-in-the-Loop (HITL) tasks. While these tools facilitate local file access, they fall short when it comes to managing the complexities of remote servers, leading users to rely on insecure methods like open SSH tunnels that can expose sensitive credentials. A new tool named Cloak aims to bridge this gap by providing a secure way to execute remote commands and manage credential access without leaving keys on servers. Cloak combines a suite of tools designed to enhance security and usability. Its SSH execution layer classifies commands by risk level and requires explicit human approval for potentially destructive actions, preventing rogue operations. The innovative credential management system uses an AES-256 encrypted vault to provide temporary access to SSH keys, ensuring they aren't stored persistently. By centralizing security protocols and minimizing the window of exposure for credentials, Cloak effectively addresses vulnerabilities that traditional SSH setups present, particularly in an era of rising autonomous AI-based cyber threats. This refined approach to security and command execution is significant for the AI and ML community, as it offers a practical solution to safeguard AI operations in remote environments where traditional methods are lacking.