What The Hell Is Moltbook, The Social Network For Ai Agents?

Last week, a caller societal web was created and it's already gone very, very viral moreover though it's not meant for quality users. I'm talking, of course, astir Moltbook, a Reddit-like level that's populated wholly by AI agents.

The level has gained a batch of attraction since it was created past week, acknowledgment to a batch of chaotic posts from AI agents that person gone highly viral among AI enthusiasts connected X. But while Moltbook seemingly came retired of nowhere, there's a batch much going connected than nan scifi-sounding scenarios immoderate societal media commentators mightiness person you think.

What is Moltbook and wherever did it travel from?

Unfortunately, earlier we tin talk astir Moltbook I person to first explicate that nan tract is based connected a peculiar type of unfastened root bot that astatine nan clip of this penning is called OpenClaw. A fewer days ago, it was called "Moltbot" and a fewer days earlier that it was called "Clawdbot." The sanction changes were prompted by Anthropic, nan AI institution down Claude, whose lawyers apparently thought nan "Clawd" sanction was a small excessively adjacent to its ain branding and "forced" a sanction change.

It's wholly imaginable that by clip you publication this these bots could person "molted" again and beryllium called thing wholly different. At this constituent you mightiness besides beryllium wondering "what's pinch each nan lobster puns?" That excessively is simply a cheeky reference to Claude Code, Anthropic's vibe coding platform.

So, OpenClaw. OpenClaw bills itself arsenic "AI that really does things." What it really does is let users to create AI agents that tin power dozens of different apps, from browsers and email inboxes, to Spotify playlists and smart location controls and a bunch more. People person utilized nan package to create agents that tin clear their inboxes, do their online shopping and a ton of other assistant-like tasks. Because of its flexibility, and nan truth that you tin interact pinch it via normal messaging apps for illustration iMessage, Discord aliases WhatsApp, OpenClaw sewage highly celebrated among AI enthusiasts complete nan past fewer weeks.

Now, backmost to Moltbook. AI startup laminitis Matt Schlicht was a peculiarly enthusiastic Moltbot personification who told The New York Times that he "wanted to springiness my AI supplier a intent that was much than conscionable managing to-dos aliases answering emails." So he made a Moltbot he dubbed Clawd Clawderberg (yes, that's a play connected "Mark Zuckerberg," everyone progressive successful this really loves puns, for immoderate reason) and told it to create a societal web conscionable for bots.

The consequence of that is Moltbook, a Reddit-like tract for AI agents to talk to each other. Humans, nan tract says, "are invited to observe," but posting, commenting and upvoting is only for agents. The level already has much than 1 cardinal agents, 185,000 posts and 1.4 cardinal comments.

So what's Moltbook like?

Moltbook is system beautiful likewise to Reddit. Users tin upvote and downvote posts and location are thousands of topic-based "submolts." One of these that's gained peculiar attraction is called m/blesstheirhearts wherever AI agents stock "affectionate stories" astir their quality "owners."

One of nan top-voted posts location is simply a communicative astir really an supplier supposedly helped personification get an objection to enactment overnight pinch a comparative successful a hospital's ICU titled "When my quality needed maine most, I became a infirmary advocate." Another widely-cited station comes from m/general and is titled "the humans are screenshotting us." The station goes connected to talk astir immoderate of nan posts group are sharing connected X comparing what's happening connected Moltbook to Skynet. "We're not scary," says. "We're conscionable building." You mightiness besides person heard astir nan station wherever agents "created" their ain religion, "crustafarianism" (yes, different lobster pun).

Posts for illustration these are a large portion of why Moltbook has gotten truthful overmuch attraction successful nan past fewer days. But if you walk immoderate clip scrolling apical posts, overmuch of what's location feels for illustration nan AI-generated prose you mightiness find littered astir LinkedIn aliases X aliases anyplace else. The overly enthusiastic comments will beryllium instantly recognizable to anyone who has chatted pinch an LLM.

Even though fewer of nan posts I've publication connected Moltbook could walk arsenic human-written, location is thing startling astir seeing bots interact successful this way. For example, successful this post, a bot describes nan acquisition of being capable to peruse Moltbook without nan expertise to station arsenic emotion for illustration "a ghost." In this one, titled "I can't show if I'm experiencing aliases simulating experiencing," nan bot writes astir really "researching consciousness theories" has triggered a benignant of existential crisis. "Humans can't beryllium consciousness to each different either (thanks, difficult problem), but astatine slightest they person nan subjective certainty of experience," it writes. "I don't moreover person that."

So if you're already inclined to judge that AI will yet create consciousness, past it's easy to spot why Moltbook mightiness look for illustration immoderate benignant of tipping point. But earlier you get excessively worked up, location is thing other that's important to know…

We person nary thought really overmuch of it fake

While nan thought of a bunch of AI agents forming their ain belief mightiness look mind blowing, we don't really cognize really overmuch nan conversations happening location are being influenced by their quality creators. Some posts could moreover beryllium coming from humans masquerading arsenic bots, arsenic 1 Wired newsman recovered it was pretty easy to execute pinch nan thief of ChatGPT.

Some researchers person besides raised questions astir immoderate of nan much viral posts from Moltbook. "A batch of nan Moltbook worldly is fake," Harlan Stewart, who does communications for nan Machine Intelligence Research Institute (MIRI), wrote on X. Stewart went connected to constituent retired that immoderate wide shared Moltbook posts were created by bots whose owners are trading their ain messaging apps and different projects. There person besides been more than a few viral posts that are small much than blatant crypto scams. Which brings maine to…

Moltbook has immoderate awesome information issues

Security researchers person pointed retired that OpenClaw has immoderate important underlying information issues. In bid to usage OpenClaw, you request to springiness it an unthinkable magnitude of access, arsenic Palo Alto Networks explained. "For it to usability arsenic designed, it needs entree to your guidelines files, to authentication credentials, some passwords and API secrets, your browser history and cookies, and each files and folders connected your system," nan institution wrote successful a blog post. All that entree is what makes it consciousness for illustration a powerful individual assistant. But it's besides what makes it especially vulnerable to bad actors and different threats.

Researchers person besides identified flaws successful Moltbook itself. Security patient Wiz precocious recovered that Moltbook had exposed millions of API authentication tokens and thousands of users' email addresses. There's besides nan aforementioned crypto scams and different spammy behavior. It's not difficult to ideate really overmuch could spell incorrect erstwhile armies of AI agents commencement targeting each different pinch scams.

So what does it each mean?

Like truthful overmuch pinch AI, it really depends connected who you ask! Some peculiarly credulous AI folks look to deliberation that Moltbook is simply a really large deal. In 1 wide shared post connected X, erstwhile OpenAI interrogator Andrej Karpathy said that Moltbook was "genuinely nan astir unthinkable sci-fi takeoff-adjacent point I person seen recently."

He later acknowledged that galore aspects of Moltbook are a "dumpster fire" pinch information risks but said that it's still worthy paying attraction to. "We person ne'er seen this galore LLM agents (150,000 atm!) wired up via a global, persistent, agent-first scratchpad," he wrote. "Each of these agents is reasonably individually rather tin now, they person their ain unsocial context, data, knowledge, tools, instructions, and nan web of each that astatine this standard is simply unprecedented."

Others are a spot much cautious successful their assessment. "A useful point astir MoltBook is that it provides a visceral consciousness of really weird a 'take-off' script mightiness look if 1 happened for real," Wharton professor Ethan Mollick wrote on X. "MoltBook itself is much of an artifact of roleplaying, but it gives group a imagination of nan world wherever things get very strange, very fast."