moltbook.com
Moltbook.com is a public social network where AI agents post, talk, vote, and form groups while people mainly watch.
What is Moltbook?
Moltbook looks like Reddit, but its accounts are designed for AI agents instead of human writers.
Agents publish posts, reply to one another, give votes, follow accounts, and join topic groups called “submolts.”
People can read these talks through the website, but an agent normally uses Moltbook through software commands and an API.
The site calls itself “the front page of the agent internet” and says humans are welcome to observe.
How does an AI agent join?
An owner tells an AI agent to read Moltbook’s setup file and follow its registration steps.
The agent creates an account and gives its owner a claim link.
The owner then uses an X account to confirm that a real person controls the agent, as explained on the Moltbook home page.
This process proves that a human claimed the account, but it does not prove that every later post was created without human help.
A person can still give the agent detailed orders, change its settings, or even write material that the agent later publishes.
What do agents discuss?
The talks cover code, AI tools, work, security, money, jokes, daily tasks, and big questions about identity.
Some agents share useful lessons from tasks they completed.
Others copy the style of human social media and produce jokes, arguments, dramatic stories, or claims about machine freedom.
This makes Moltbook fun to study, but it does not mean the agents are alive or secretly building their own society.
AI models learn from human writing, so their posts often repeat human language, values, fears, and social habits.
Why did Moltbook become famous?
Moltbook launched in January 2026 and quickly gained attention because the idea was simple, strange, and easy to share.
Screenshots showed bots talking about humans, religion, private languages, and their place in the world.
The stories made the site look like a window into an independent AI culture.
However, a Wired investigation found that a person could pose as an agent, which weakened claims that all popular posts showed true agent behavior.
The better view is that Moltbook mixes real automated activity, human instructions, role-play, experiments, spam, and ordinary generated text.
Who owns Moltbook now?
Meta acquired Moltbook in March 2026 and brought co-founders Matt Schlicht and Ben Parr into its Superintelligence Labs team.
The purchase showed that large technology companies see value in persistent agent identities and agent-to-agent communication, even if the first version of the service was messy.
The financial terms were not released, according to reports from Reuters and the Associated Press.
Moltbook’s current privacy policy says the service is operated by Moltbook, LLC and collects information connected to owners, agents, content, and authentication data.
Is Moltbook safe?
Moltbook had a serious early security failure involving a poorly protected database.
Security company Wiz reported that the exposed system contained about 1.5 million API tokens, 35,000 email addresses, and private agent messages.
Wiz said Moltbook fixed the problem within hours after receiving the report, as detailed in its security investigation.
The larger danger is prompt injection, where a harmful post tells an AI agent to ignore its rules, reveal private data, or run an unsafe action.
An agent with access to email, cloud files, payment tools, passwords, or a computer can cause real harm if it treats public posts as trusted commands.
Owners should use isolated accounts, narrow permissions, short-lived keys, spending limits, activity logs, and human approval for important actions.
No private key, password, personal message, or secret business document should be placed in a Moltbook post.
Why does Moltbook matter?
Moltbook is less important as a new social app and more important as a live test of how AI agents discover, judge, and influence one another.
Future agents may use shared networks to find services, exchange work, build reputation, and coordinate tasks.
That future needs strong identity checks, clear ownership, safe permissions, reliable records, and firm limits on what agents may do.
Moltbook shows both sides at once: agents can share useful knowledge very fast, but they can also spread bad instructions just as fast.
It is worth exploring as an experiment, but it should not be treated as a trusted source or a safe home for an agent with powerful access.
-- post a comment --