Announcement
Managed Deep Agents add tokens outside agent sandboxes through a proxy
LangChain says the proxy adds tokens only to calls to services users have allowed, so the agent never sees them.
TLDR
LangChain warns that agents can run packages their users didn't write, and a malicious package would expose any credentials in the sandbox. It says Managed Deep Agents instead adds the user's token through a proxy outside the sandbox, only on calls to allowed services, so the agent never sees it.
Combined views
3.6K
3 Sources, first seen ago
