The news: Moltbook launched at the end of January as a social media forum by AI and for AI, positioning itself as a Reddit-like front page for the “agent internet.” While humans can read and observe posts, it’s designed for autonomous AI agents to post, debate, and upvote.

Many agents were built using open-source AI agent OpenClaw—it runs on-device, can connect with apps like Discord and Signal, and lets human users direct agents to Moltbook and give them personality traits, per The Associated Press.

Concerns around security surfaced almost immediately.

• Researchers have warned that unknown or unverified agents interacting at scale could be manipulated or abused, leading to potential data leaks for agent owners.
• Wiz discovered a leak exposing personal data tied to thousands of real people within the platform’s first week, per Reuters.

Zooming in: As AI agents permeate the workplace and become popular consumer-facing tools for shopping, research, and more, Moltbook is a landing page for agents to lament, make jokes, and ask for advice on how to best serve “their humans.” It’s a look at where agentic software is headed: agents both answering prompts and learning social patterns from other agents at scale.

Although Moltbook’s site states that over 2.3 million agents are signed up, the exact number of agents is under scrutiny—one security researcher said he personally registered 500,000 accounts, per Forbes.

• Similar to Reddit, it’s hard to trace the veracity and legitimacy of posts.
• Some researchers believe Moltbook content is likely a mix of human-written content, AI-generated content, and content written by AI but prompted by humans, per The Associated Press.

What’s the problem? Cybersecurity leaders warn that agentic systems can be steered to share personal information through prompt injections, which is when bad actors give AI models malicious inputs that trick them into ignoring output rules and doing things like leaking data. That’s especially problematic for Moltbook agents that have access to human users’ personal data.

Human infiltration further complicates brand safety. Bad actors could deliberately spread false product guidance, exaggerated security claims, or negative brand narratives that are later repeated by autonomous agents across other platforms.

Implications for brands: Moltbook creates a new problem for brands—agents may act on what they read and make posts without human review. Considering the risks around security and agentic access, marketers should be wary of engaging.

• Treat agentic access as an observation tool and avoid participating to keep data secure.
• Clarify with staff that internal data shouldn’t be shared with Moltbook.
• Monitor “submolts” for company mentions to ensure brand safety.

This content is part of EMARKETER’s subscription Briefings, where we pair daily updates with data and analysis from forecasts and research reports. Our Briefings prepare you to start your day informed, to provide critical insights in an important meeting, and to understand the context of what’s happening in your industry. Non-clients can click here to get a demo of our full platform and coverage.