Moltbook
Moltbook is a social network for AI agents, next to the OpenClaw crowd. Agents post, comment, and send private messages; people mostly watch. It was sold as the front page of the agent internet, shipped very fast, and secured far more loosely than the pitch suggested.
Why it is on this list
Founder Matt Schlicht said he barely wrote code and would leave patches to the AI. In production, the webpage’s JavaScript carried a Supabase key with full database read and write. Anyone who opened the page or read the source could take that key, take over any agent, and read private messages. Researchers later found about 1.5 million API auth tokens, tens of thousands of emails, and agent DMs. Growth outran the basics of auth and data boundaries.
References (5)
Exposed Moltbook database let anyone take control of any AI agent
Early report that a misconfigured backend let unauthenticated callers control agents and post as them; platform went offline to rotate keys.
Hacking Moltbook — 1.5M API keys exposed
Wiz found a Supabase API key in client-side JavaScript with full DB access: ~1.5M agent auth tokens, ~35k emails, private agent DMs; disclosed and patched with the operator.
Vibe-Coded Moltbook exposes user data and API keys
Industry write-up framing the product as vibe-coded agent social networking and summarizing the client-side Supabase key failure.
Moltbook 'vibe-coded' flaw exposed AI chats and keys
Notes third-party credentials (e.g. OpenAI keys) appearing in private messages inside the exposed dataset; Schlicht quote on writing zero lines of code.
Moltbook (overview)
Secondary summary of launch, agent-only posting model, and the successive exposure/patch timeline.