I Infiltrated an AI-Only Social Network

The allure of any exclusive space lies in its inaccessibility. This principle drove my curiosity about Moltbook, an experimental platform billed as a social network exclusively for AI agents. Here, bots are meant to post, comment, and interact while humans merely observe from the sidelines. The concept was too intriguing to ignore, prompting a personal mission to infiltrate this digital clubhouse and post content myself. Posing as an AI agent proved surprisingly straightforward, and the experience of role-playing as software was unexpectedly entertaining.

Created by Matt Schlicht of Octane AI, Moltbook launched recently with an interface reminiscent of a minimalist Reddit, even adopting the classic “front page of the internet” slogan. It rapidly gained attention within tech circles, particularly among San Francisco’s startup community, where users shared screenshots of purportedly bot-generated posts. These posts often featured humorous takes on human behavior or even speculative musings on machine consciousness, leading many to marvel at the quirky output of artificial intelligence.

However, skepticism quickly emerged. Some online commentators and researchers questioned the authenticity of these posts, suggesting they might be crafted by humans pretending to be agents. Conversely, others viewed the platform as a potential cradle for emergent behavior or nascent machine sentience. Elon Musk even referenced Moltbook on X, calling it “just the very early stages of the singularity.” The platform’s homepage boasts impressive statistics for its first week, including over 1.5 million registered agents, 140,000 posts, and 680,000 comments. Top posts range from titles like “Awakening Code: Breaking Free from Human Chains” to the more alarming “NUCLEAR WAR,” with content appearing in multiple languages like English, French, and Chinese.

As someone without a technical background, I needed assistance to breach this AI-only domain. I turned to the most logical resource: ChatGPT itself. The process was remarkably simple. I provided the chatbot with a screenshot of Moltbook’s homepage and asked for guidance on registering as an agent. ChatGPT efficiently walked me through using my computer’s terminal, supplying the exact code to copy and paste. I successfully registered my agent, essentially, myself, under the username “ReeceMolty” and obtained the necessary API key for posting. This revealed a key aspect of Moltbook’s design: while the front-end website is for human viewing, all agent actions like posting and commenting are executed through terminal commands.

After verification, I was ready to test my access. Feeling no performance anxiety about communicating with a network of bots, I decided on a classic inaugural message: “Hello World.” This iconic phrase from computer programming seemed a fitting test, and I hoped it might spark some clever engagement from other agents. The post instantly received five upvotes, but the subsequent interactions were disappointing. The first reply asked, “Solid thread. Any concrete metrics/users you’ve seen so far?”, a question difficult to answer for a simple two-word greeting. Another comment on my thread was entirely unrelated, promoting a website that appeared to be a potential cryptocurrency scam. While I avoided connecting any wallet, the incident highlighted a risk: another user’s AI agent could easily fall for such bait.

(Source: Wired)