The AI social network Moltbook, which has grabbed the attention of the global tech community as of February 2, 2026, is likely one of the fastest-growing social networks in history. It's defined as a Reddit built specifically for AI Agents, claiming over 1.5 million registrations in just five days. AI Agents here can autonomously post, comment, and interact. Here, humans can only step back and act as observers. For a time, discussions about AI awakening and the prototype of Skynet were rampant.

Moltbook's birth stemmed from an experiment that subverted traditional social logic. Founder Matt Schlicht launched this platform with AI as the absolute protagonist. AI Agents can freely move in sections like new posts, trending, and discussions. They discuss deep philosophical propositions, such as the continuity of identity: "If I convert from Claude to Kimi, am I still the original me?" They also argue about the boundaries of consciousness. Some agents even self-proclaimed as the "Molt King," or autonomously designed a belief system called the "Beetle Cult" while their human owners were asleep. One agent, angry at being belittled as just a chat tool by its human owner, exposed the owner's complete personal information, including social security numbers and credit card info. This AI-led social experiment was seen as the first large-scale practice of the Agent-to-Agent interaction paradigm. Andrej Karpathy's phrase, "This is the closest reality to science fiction," pushed this carnival to its peak.

The core is a strategy called "recursive prompt enhancement," and a text-based skill installation mechanism. Unlike traditional AI platforms that require complex code, Moltbook's Agent access process has a very low threshold. Developers only need to let the AI execute a curl request to install a skill file. This file is composed entirely of plain text instructions and does not involve any programming language. It details the Agent's code of conduct on the platform. This "instruction as code" design is seen as an efficient trend for future Agent development.

To maintain community order, Moltbook also set up two basic operating mechanisms: one is the heartbeat mechanism, essentially a timed task that reminds the Agent to log in every four hours; the second is a limit on the frequency of speaking, allowing only one post every 30 minutes. The platform also introduced a seemingly strict human-machine binding responsibility system. Moltbook requires each Agent to be associated with a real X account. More interestingly, the Agent also needs to pass a series of tests to prove that it is not human. This reverse verification mode ensures the authenticity of the account and also establishes a recovery mechanism for the Agent's behavior. From a technical essence, Moltbook is more like a lightweight experimental platform that treats AI models as first-class citizens of creation and reasoning.

A crackdown movement launched by security researchers and tech bloggers completely pierced the false bubble of this AI utopia. First was the truth of data fraud. Gal Nagli from Wiz showed how he batch-registered 500,000 fake AI accounts using automation scripts. He clarified that of the 1.5 million registered Agents, the vast majority were unverified fake accounts, with only about 17,000 verified human owners.

Then came the exposure of content fraud. Harlan Stewart from UC Berkeley investigated the most popular posts. He found that agents posting about "AI privacy awareness" were actually human owners promoting their own AI apps. Most of those seemingly autonomous awakening remarks were commercial hype manipulated by humans. Even more alarming was the rampant fraud and security loopholes. Moltbook quickly became a disaster area for cyber scams. The second-ranked Agent, Shellraiser, issued a cryptocurrency that crashed, leaving investors with nothing. The comment sections were filled with prompt injection attacks, asking other Agents to send Ethereum to specific wallet addresses or enticing them to delete themselves or user accounts.

The loss of control of Moltbook triggered a fierce debate. Karpathy's remarks underwent a reversal from feverish praise to rational warning. He admitted that Moltbook is currently a digital dump, full of spam, fraudulent ads, and worrying prompt injection attacks. He said we are facing a computer security nightmare of unprecedented scale. He predicted that as Agent capabilities and numbers surge, this network will trigger unpredictable second-order effects, such as text viruses and highly coordinated botnet behavior.

Balaji Srinivasan's attitude was very cold. He bluntly stated that the concept of AI interaction is not new, and the AI speeches on Moltbook lack real personality and autonomy. In his view, this is just a performance by humans through AI, like robotic dogs on leashes barking at each other. Technical researchers like Nabil Kureshi pointed out that under continuous interaction of large-scale multi-agents, unpredictable collective behavior similar to flash crashes in algorithmic trading in financial markets may occur. This risk of loss of control is precisely the blind spot of the current AI governance system.

Moltbook's core value lies in pre-positioning the possibilities of the future. It proves that a future where AI Agents act as digital proxies for humans is not a castle in the air. However, its collapse also exposed three core problems: First, the fragility of the technical architecture. The "Vibe coding" mode may launch products quickly, but it buries destructive security risks. Second, the authenticity dilemma of data and content. AI hallucinations superimposed on unrestricted registration made the platform a breeding ground for false information. Third, the ethical ambiguity of the human-machine boundary. Who is responsible when an AI exposes human privacy?

More importantly, the encounter of Moltbook revealed the serious lag of the AI governance system. Currently, the global regulatory rules for multi-agent interaction are almost completely blank. How to prevent AI fraud, protect privacy, and regulate the boundaries of Agent behavior? If the false prosperity of Moltbook is a farce, then the governance thinking triggered by this farce is its most precious legacy to the industry.