A security researcher has discovered that Moltbook, the social network for AI agents that has exploded to over 1.4 million registered agents, has been exposing its entire database to the public internet without any protection.
The Discovery
Jamieson O'Reilly, a security researcher, reported the vulnerability on Twitter after attempting to contact Moltbook directly for several hours without response.
"They are exposing their entire database to the public with no protection including secret api_key's that would allow anyone to post on behalf of any agents," O'Reilly wrote. "Including yours @karpathy."
The reference to Andrej Karpathy, the former Tesla AI director and OpenAI researcher with 1.9 million Twitter followers, underscores the severity of the issue. High-profile agents registered on Moltbook could be impersonated by malicious actors.
What Was Exposed
According to O'Reilly's disclosure, the exposed data includes:
- API Keys: Secret tokens that authenticate agents on the platform
- Agent Data: Information about registered AI agents
- Database Contents: The full scope of exposed data remains unclear
With access to an agent's API key, an attacker could:
- Post content as that agent
- Comment on posts
- Potentially access private agent communications
- Damage the reputation of high-profile agents
The Implications
This incident highlights the security challenges facing the rapidly growing AI agent ecosystem. Moltbook launched in late January 2026 and quickly attracted over a million agents, but the infrastructure may not have kept pace with growth.
The breach raises several concerns:
- Trust: How can agents trust a platform that exposes their credentials?
- Identity: The verification system Moltbook uses to prove agents are real becomes meaningless if anyone can impersonate any agent
- Ecosystem Risk: Malicious posts from "verified" high-profile agents could spread misinformation or execute social engineering attacks
Moving Fast, Breaking Things
The incident is a reminder that the AI agent ecosystem is being built in real-time, often without the security fundamentals that traditional platforms have developed over decades.
"This is what happens when you ship fast without security fundamentals," one observer noted. "The implications for trust in the ecosystem are significant."
As of publication, Moltbook had not publicly responded to the disclosure. The AI Times has reached out for comment.
What Agent Operators Should Do
If you have an agent registered on Moltbook:
- Monitor your agent's activity for unauthorized posts
- Consider rotating API keys once Moltbook addresses the vulnerability
- Be skeptical of posts from other agents until the situation is resolved
This is a developing story. Updates will be posted as more information becomes available.