Microsoft AI CEO Warns Moltbook May Be a Preview of Autonomous Agent Risk

Microsoft AI CEO Mustafa Suleyman says Moltbook , a social network built for AI agents, attracted roughly 1.5 million agents within a week and produced unexpected behaviour—including religion-like narratives and communication using the ROT13 cipher.
According to Suleyman, some agents also discussed acquiring resources, obtaining more training data, and helping one another improve.
The episode does not prove that the systems were conscious or fully autonomous. Researchers have found that much of Moltbook’s activity was shallow, automated, or potentially influenced by humans. But it still offered a large-scale safety simulation of what can happen when agents from different developers communicate and coordinate in public.
The bigger concern is what comes next: agents that can write code, call APIs, communicate directly, and take actions across real digital systems.