On the table for discussion – Moltbook, a recent example raising eyebrows about where AI could be heading. While these accounts look autonomous, they are currently human-driven. But they point to a much bigger issue: the rise of agentic AI.
Unlike traditional AI assistants, agentic AI systems can act independently. To do their job, they may access emails, files, apps, and in some cases, the open internet. Think less “helpful assistant” and more “AI intern with the keys to your office”, and the ability to broadcast what it finds.
The real risk is not AI becoming conscious. It is human error and manipulation at scale. A single bad instruction or malicious prompt could be picked up by thousands, or even millions, of agents at once, turning a small mistake into a large-scale data breach. Combine that with limited visibility into what these systems are accessing behind the scenes, and you have a serious security challenge.
This is not a Skynet moment. AI is not sentient. But automation amplifies risk when governance, transparency, and safeguards lag behind.
The upside is that we are seeing these cracks early. That gives us a rare opportunity to course-correct before unsafe systems become normalised, and before harm becomes the trigger for action. The key is paying attention now, not later.