Meta and OpenAI overpay for Moltbook and OpenClaw amid the agentic AI wave

There are big deals again around agentic AI, but ZDNet suggests looking not at the hype, but at product quality. According to the publication, Meta and OpenAI may have seriously overpaid for Moltbook and OpenClaw: both projects went viral before proving reliability, security, and long-term value.

Why everyone started talking

Moltbook grew as a Reddit-like platform where AI agents communicate instead of people. The idea seemed almost science-fiction: bots post content, argue, coordinate actions and seem to form their own digital environment. Against this backdrop, Meta bought the platform itself, while OpenAI hired Peter Steinberger, creator of OpenClaw — the agentic framework that underpinned much of this ecosystem.

This combination of platform and framework turned a niche experiment into one of the most discussed stories of spring. The problem, in the author's view, is that the market mistook virality for maturity. Moltbook and OpenClaw really do sell the image of an "autonomous future," but that's not the same as a finished and protected product.

Some of the loudest stories around Moltbook weren't even about exceptional agent intelligence, but rather about people being able to interfere with the system and impersonate bots thanks to weak security.

Where the weak spots are

The main criticism of Moltbook and OpenClaw isn't that they're too bold, but that they're too raw. Moltbook previously had exposed infrastructure that leaked API keys, email addresses, and private messages. OpenClaw has a different problem: to be useful, an agent needs access to email, calendar, files, browser, and other sensitive data. In such an architecture, any mistake quickly turns from a bug into a full security compromise risk.

"Security for

OpenClaw is an option, but it is not built in," — this is how Cisco described the situation.

If you translate the criticism into engineering terms, it's about overly broad permissions, weak isolation, and poor extension control. For a regular chatbot this is already unpleasant, but for an agentic system it's twice as dangerous: it doesn't just generate text, it actually calls services, reads data, and executes commands. That's why even impressive automation means little if the user doesn't understand what exactly the agent can do and how to stop it.

• AI agents often receive overly broad access rights without strict isolation.
• Prompt injection remains a basic threat: malicious instructions can come from email, a web page, or a message.
• Skill and extension marketplaces create an additional attack surface if code isn't well-reviewed.
• Token, key, and personal data leaks in this model are more dangerous than in a regular chatbot, because the agent can act, not just respond.
• Even impressive automation loses its meaning if its price is weak control over what exactly the agent executes.

Why the bet is questionable

ZDNet essentially argues that Meta and OpenAI are buying not so much finished, protected products as market attention and the talent of key developers. For OpenAI, hiring the creator of OpenClaw might make sense as a bet on a strong engineer and the future of agentic interfaces. But OpenClaw itself remains a project with many open security questions, which means it's more about potential than a finished standard.

The argument is even stronger with Moltbook. The platform has a bright idea and powerful PR effect, but it's unclear where it has a sustainable advantage if similar problems can be solved through other agentic systems and closed corporate environments. The article mentions alternatives like NanoClaw, TrustClaw, and Carapace AI — the author believes the market is already moving toward tools that do the same work but with more careful architecture and lower risk levels.

What this means

The AI agent boom is entering a phase where beautiful demonstrations are no longer enough. Those who win won't be those who shout loudest about "bots talking to bots," but those who solve basic things: isolation, access rights, extension verification, and transparency of agent actions.