Exposed Clawdbot/Moltbot agent control panels leak keys and chats
A default trust setting meant that common hosting setups gave strangers full access to users' personal AI agents and the accounts connected to them.
- What happened
- Clawdbot, an open-source personal AI agent that grew quickly in January 2026, runs a gateway and control panel that trusts connections from localhost. When users placed it behind reverse proxies such as Nginx or Caddy, internet traffic arrived as localhost traffic and was approved automatically. Researcher Jamieson O'Reilly of Dvuln found hundreds of internet-facing control interfaces through Shodan, some fully open without authentication. Exposed data included API keys, OAuth tokens, SSH private keys and full conversation histories. Agent memory and credentials were also stored in plaintext on disk.
- Response
- The gateway authentication bypass was patched. The project renamed itself Moltbot after a trademark request, and later OpenClaw. VentureBeat reported that RedLine, Lumma and Vidar infostealers added Clawdbot paths to their target lists within 48 hours.
- Discovered by
- Jamieson O'Reilly (Dvuln), using Shodan; Hudson Rock later reported infostealer malware families adding Clawdbot data paths to their targets.
- Who was affected
- Hundreds of internet-facing agent deployments; API keys, OAuth tokens, SSH keys and months of private chat history
- Patterns
- Sources
Related
Incidents with the same pattern
What does this ask of people?
Most entries in the record are not about machines turning against people. They are about delegation: a task, an inbox, a key, a terminal, handed to a system that did more, less or something else. Futures literacy starts with noticing that.