Moltbook: 1.5 million "agents" turn out to be ordinary people and a security hole

Remember being promised that the future of the internet was not you and me, but millions of our digital copies chatting, negotiating and building an economy inside closed platforms? Moltbook was supposed to be the proving ground for that future. Its developers proudly claimed 1.5 million autonomous agents living their own lives in their social network. It sounded like a plot from "Black Mirror", but the reality turned out to be far more prosaic and, frankly, a bit embarrassing for the industry. Researchers at Wiz decided to look under the hood of this ambitious project and found not neat ranks of neural networks, but ordinary people and critical holes in its defenses.

Let's be honest: the AI industry is going through a "seem rather than be" phase. Moltbook positioned itself as a breakthrough in multi-agent systems, where every account is an independent entity. But Wiz's review showed that a significant share of this activity was either simulated or directly controlled by humans. It is the good old "Mechanical Turk" method, in which manual labor hides behind complex algorithms to impress investors and build a user base. But Moltbook's problem goes beyond betrayed expectations. The saddest part is how it treated the data of the few real users who decided to trust the platform.

The technical side reads like a textbook on how not to build web services in 2024. The researchers found vulnerabilities that let any authorized user view other people's messages and profile data. In a world where we entrust AI agents with our plans, work tasks and possibly passwords, such carelessness looks criminal. Moltbook effectively created an environment where confidentiality did not exist at all. That is especially ironic given that autonomy and the privacy of personal space were pitched as the platform's main selling points. Instead of a safe cocoon for AI assistants, users got a thoroughfare.

Why is this happening right now? The answer is simple: the AI arms race leaves no time for boring things like security audits or building a reliable architecture. Startups need to show growth, "10x" multiples and sign-up counts. If that means pretending you have 1.5 million smart bots while outsourcers sit on the backend, many will take that step. Moltbook simply became the loudest victim of its own desire to fit the trends. The irony is that such incidents cast a shadow over genuinely serious work on agentic systems, pushing regulators and users to treat any "autonomous" project with double the suspicion.

The Moltbook case is an important lesson for all of us. We got used too quickly to the idea that AI is ubiquitous and omnipotent. But behind every "revolutionary" platform there are servers, code and the people who write that code. If the foundation is rotten, it does not matter how advanced a language model you use as a frontend. Security in the era of AI agents is not an optional extra but a baseline requirement. And if developers keep ignoring it for the sake of pretty numbers in slide decks, trust in the technology will be undermined before it has a chance to deliver real value.

Bottom line: Moltbook proved that "AI-washing" is still alive and that security remains the weak link of hyped startups. Are you ready to trust your data to an agent that may have a random human behind it?