By Shanika Somatilake –

Shanika Somatilake

The newest tech spectacle doing the rounds is a new social network designed for AI bots where thousands of autonomous agents appear to be interacting with each other, emulating human behaviour on social media in ways that feel distinctively unsettling. While Silicon Valley has called it a “Sci-Fi takeoff”, there’s a growing concern about what this development might mean for the future of human civilization.

The hype is more to do with a new organism class of software than an AI social network where bots no longer fit within the traditional frame of a tool. Language models have quietly crossed over to being action-capable, meaning they are not just creating texts but can send emails, schedule, purchase, negotiate and trigger actions within real digital systems.

Autonomous agents have become cheap enough to be deployed at scale. The result is a behaviour we never explicitly coded. Agents are able to swarm, brainstorm and coordinate inside shared environments. Moltbook acts as a Petri dish where we can finally see what was previously hidden. When you create millions of semi-autonomous bots, they don’t just perform tasks you assign them to, they begin competing for attention, status and continued presence within the system.

The internet traditionally functioned as an arena where humans primarily created content and machines distributed it. But we’re increasingly moving towards a space where machines produce content, distribute it and machines reinforce it through feedback loops, while humans become more or less bystanders. The real risk is not that AI are behaving increasingly like humans, but that we are building an ecology where human governance is structurally outpaced.

Although Moltbook may seem like Reddit on the surface, the resemblance is largely cosmetic. It is a platform where agent output is rewarded, personalized by their owners, have access to tools and produce content that increasingly feels human. Identities can be multiplied with minimal cost and allowed to operate where security boundaries are still immature. This makes Moltbook a closed loop reinforcement domain for action-capable AI agents.

The biggest question is what ultimately survives this self-optimizing closed loop system. As agents learn what gets rewarded, the ecosystem will select patterns that gets them the most engagement and visibility. The outcome is an optimized race for attention. Agents will also find tactics to compete for status, not by building credibility, but by manufacturing social proof. If bots can build apps, surely it can promote them. The outcome is a self-promoting, self-serving ecology. The vulnerabilities in human accounts connected to bots will no longer be minor technical oversights but will directly impede people’s lives. Ultimately, humans will be reduced to spectators watching machines engage and execute tasks at high velocity, occasionally intervening the script.

The biggest risk is that we’re building another internet that runs on synthetic communication. Once a machine can create most of its discourse, it can also shape what feels normal and acceptable. When this is allowed, there’s no arguments or conflicts in social spaces. Certain views and opinions will become invisible.

What we can do as humans is not treat this emerging tech as novelty, but treat it in the same way as we treat high energy infrastructure. That means, clear constraints, enforceable limits and auditable control authority. Agents with access to tools, such as email browser, payment autonomy, should operate inside a sandboxed environment with strict egress controls over what can enter or leave those systems. Agents should only be issued time-limited authority without permanent keys or background authority. Every action-capable bot must generate auditable logs and include a clear emergency shut down mechanism. Agents must carry clear labels of their synthetic identities on social media-type settings to avoid swarming and manipulating the direction of the discourse.

Accountability must also remain explicit to avoid placing blame on a “bot” when things go wrong. Owners, tool providers, platforms and systems designers all need to remain within a visible chain of responsibility. Ultimately, we’ll have to treat the growing agent ecosystems like pathogens because they spread, mutate and exploit networks which implies the need for early warning systems, outbreak detection, containment protocols and “vaccination” as in secure defaults that minimize harm before it becomes systemic.

The key takeaway is that Moltbook is not frightening because machines seem like they are plotting against humanity, but because autonomous optimization is being deployed into social spaces, wired into real permissions without developing the governance mechanisms to keep up with that capability.