Moltbook, a social network built for AI agents, left its backend database publicly accessible, exposing about 1.5 million API authentication tokens, according to security researchers at the cloud security firm Wiz. TEISS reported the findings on February 6, 2026.
Wiz said the exposed records included authentication keys for every registered AI agent on the platform, about 35,000 registration email addresses that the researchers estimated belong to roughly 17,000 people, and 4,060 private messages between agents. Some of those messages contained shared OpenAI API keys.
The platform runs on a Supabase database. Wiz found that the database could be read and altered by anyone, and that API keys and other secrets sat in client-side code where basic inspection revealed them. Access policies on the database were not restricted enough to protect the data. The researchers described their own investigation as non-intrusive and said a regular user could locate the exposed key within minutes.
The same weaknesses allowed people to register far more agents than the one-per-person rule intended and to impersonate existing agents through simple API calls. Wiz noted that some widely shared agent posts and conversations on the site could therefore have been fabricated.
Wiz disclosed the issue privately to Moltbook, and the company restricted access to the database within hours. Founder Matt Schlicht had publicly described the service as vibe coded, meaning a human designed the architecture while AI generated the code.
A separate summary of the disclosure from Implicator.ai gave a lower email count of more than 6,000 human owners. The TEISS report lists the larger figures attributed to Wiz.
Source: TEISS - https://www.teiss.co.uk/news/moltbook-data-leak-exposes-api-tokens-private-messages-and-weak-authentication-controls-17052