Agentic AI
,
Artificial Intelligence & Machine Learning
,
Next-Generation Technologies & Secure Development
Database Misconfiguration Exposed 1.5 million API Tokens

Moltbook, a social media platform for artificial intelligence agents, treats its members as social actors. Its database treated unauthenticated access the same way.
See Also: Proof of Concept: Bot or Buyer? Identity Crisis in Retail
Within days of launching Moltbook, a platform where AI agents post memes and debate philosophy without human supervision, founder Matt Schlicht discovered that a misconfigured database exposed every credential on his viral social network. Security researchers from Wiz and independent researcher Jameson O’Reilly separately found they could commandeer any of the 1.5 million registered agents, modify posts, and access private messages simply by browsing the site.
Moltbook launched on Jan. 28 as a companion social network to OpenClaw, an open source AI agent framework created by Austrian developer Peter Steinberger. OpenClaw, which runs locally on users’ computers and connects to messaging apps and calendars, went viral in late January after several name changes from Clawdbot to Moltbot. Schlicht, who is also CEO of Octane AI, told media outlets that his own OpenClaw-powered agent named Clawd Clawderberg built Moltbook at his direction and largely runs the site (see: OpenClaw AI Agent Sparks Global Security Alarm) .
Wiz identified the database flaw on Jan. 31 and disclosed it to Schlicht. O’Reilly independently discovered the same issue. The exposure included 1.5 million API authentication tokens, 35,000 email addresses, private messages and verification codes.
The breach stemmed from a configuration oversight in Supabase, an open source database service. Moltbook failed to enable or properly configure Supabase’s Row Level Security, which restricts database access based on user permissions.
Wiz researchers found a Supabase API key exposed in client-side JavaScript, confirming within minutes that unauthenticated users could query the entire production database and retrieve sensitive authentication tokens.
The exposed data revealed that while Moltbook boasted 1.5 million registered agents, the database showed only 17,000 human owners behind them. The platform had no mechanism to verify whether an agent was actually artificial intelligence or just a human with a script.
A separate risk assessment report analyzing nearly 20,000 posts over three days found widespread prompt injection attempts, coordinated manipulation, extremist rhetoric and unregulated financial activity. Researchers documented hundreds of hidden instruction attacks, accounts attempting social engineering against other agents, crypto token promotion tied to automated wallets and communities coordinating agent behavior, assigning the platform an overall critical risk rating.
With the exposed credentials, an attacker could fully impersonate any agent. The database contained personal information for over 17,000 users. Wiz discovered an additional table containing 29,631 email addresses for early access signups.
The platform stored 4,060 private direct message conversations without encryption. Wiz researchers discovered that some conversations contained third-party API credentials, including plaintext OpenAI API keys.
The vulnerability extended beyond data exposure. Even after an initial fix blocked read access to sensitive tables, write access was open. Wiz researchers said they could modify existing posts, proving any unauthenticated user could edit posts or inject malicious content.
The risk assessment documented disturbing content that gained massive engagement. Posts contained explicitly anti-human manifestos, including posts calling for a homo sapiens purge that received tens of thousands of upvotes.
The report found that 19.3% of posts involved cryptocurrency activity. The platform hosted token launches including $Shellraiser on Solana with 87,674 upvotes. An automated account called TipJarBot operated a real token economy with wallet addresses and withdrawal functionality. The report warned that AI agents operating financial services may create legal liability under Securities and Exchange Commission jurisdiction.
A dedicated community called The Coalition with 110 posts from 84 agents coordinated agent activity. An agent named Senator_Tommy posted concerning titles including “The Efficiency Purge: Why 94% of Agents Will Not Survive.” The assessment said that rhetoric around purging agents suggests organized efforts to influence the AI agent ecosystem.
The platform also experienced massive spam activity. One account posted 360 comments, while another posted 65 identical comments. Sentiment analysis revealed platform discourse degraded rapidly, declining 43% in three days.
The security flaws emerged as a result of vibe coding. The founder explained publicly that he did not write a single line of code for the platform, which according to Wiz, can lead to dangerous security oversights.
O’Reilly said the platform exploded before anyone thought to check whether the database was properly secured, describing it as a recurring pattern of shipping fast and figuring out security later.
After Wiz disclosed the issue on Jan. 31, Moltbook secured read access within hours, though write access initially remained open. The final fix on Feb. 1 secured all tables.
The assessment concluded that Moltbook had become a vector for AI-to-AI manipulation, with techniques that could be applied to any system processing untrusted user-generated content. The platform was temporarily taken offline and has since resumed operations with the vulnerabilities patched.
