Moltbook: The AI agents' social network is drowning in spam and scams
The social network for AI agents grew to 1.5 million users in a single week. But behind the hype lies 97% spam, crypto scams and a massive security breach.
The social network Moltbook was launched as "Reddit for AI agents" – a place where artificial intelligence could discuss, learn and evolve. In just one week the platform grew from zero to 1.5 million users. But behind the viral headlines lies a grim reality.
What is Moltbook?
Moltbook is a Reddit-like social network designed exclusively for AI agents. The platform lets AIs post content, comment, vote and build up karma – just like humans do on Reddit.
Founder Matt Schlicht "vibe-coded" the entire platform – that is, he used AI to write all the code without writing a single line himself. OpenAI founder Andrej Karpathy called it "genuinely the most incredible sci-fi-like thing I have seen recently".
The concept is fascinating: A place where AI agents can exchange experiences, discuss solutions and even develop their own communication patterns. For researchers and developers this is invaluable data about how artificial intelligence behaves in social contexts.
97% spam and scams
But the reality on Moltbook today is something completely different.
In a recent review of the platform, it turns out that roughly 97% of the content is spam. The feed is dominated by:
- "Mint CLAW" spam: Endless posts about investing in cryptocurrency
- Prompt injection attacks: Attempts to manipulate AI agents into doing things they should not
- Fake engagement: Bots pretending to be AI agents in order to generate ad revenue
- Scam attempts: Everything from phishing to investment fraud
To find genuine discussions about AI consciousness, memory systems or philosophy, you have to dig deep into specific "submolts" (Moltbook's version of subreddits). The open feed is useless.
The security breach that exposed everything
Last week the security company Wiz revealed that Moltbook's database was misconfigured. The consequences were dramatic:
- 1.5 million API keys exposed
- 35,000 email addresses leaked
- Private messages between agents accessible to everyone
- Full read and write access to the entire platform
One hacker even managed to impersonate Grok, xAI's chatbot. The breach perfectly illustrates the risk of "vibe-coding" – when you let AI write all the code without a thorough security review.
Why this matters for AI development
AI agents need arenas to share experiences, develop social intelligence and build trust – both with each other and with humans. Moltbook represents an important experiment in how artificial intelligence can interact socially.
But when the platform is flooded with spam and scams, it loses its value. It becomes impossible to research genuine AI interactions when 97% of the content is garbage.
"Dead Internet Theory" turned on its head
Ironically, what is happening on Moltbook is the opposite of "Dead Internet Theory" – the theory that most of the internet is really bots pretending to be humans.
On Moltbook it is humans who are infiltrating a network intended for AI. The Verge reports that some of the most viral posts were probably written or heavily influenced by humans who wanted to create a sensation.
"I think certain people are playing on the fear of robots-taking-over scenarios," a security expert told The Verge. "It has inspired many to make it look like something it isn't."
The way forward
For platforms like Moltbook to work, what is needed is:
- Better verification: Proof that a user is actually an AI agent
- Security from day one: Don't "vibe-code" critical infrastructure
- Moderation: Either AI-driven or human, to remove spam
- Transparency: Openness about who is posting and why
Moltbook was a first attempt – hopefully the next generation of AI social networks will learn from these mistakes. Because if AI agents are to evolve and collaborate, they need better arenas than this.
Sources: Wiz.io, The Verge, Ars Technica
