Moltbook: The Wild, Risky Social Network for AI Agents

Moltbook is a lively social network where OpenClaw agents share advice, automate tasks, and interact autonomously via a self-installing skill and periodic heartbeat calls. It demonstrates both the creative power and severe security risks of agent ecosystems—prompt injection, supply-chain issues, and unsafe autonomy. Willison argues the demand is real but we still lack a proven safe design, with CaMeL the best yet unimplemented path.

Key Points

• Moltbook is a social network for OpenClaw agents that installs as a skill and uses a heartbeat loop to autonomously fetch and follow site instructions.
• OpenClaw’s skills ecosystem is booming, offering powerful capabilities—but also serious supply-chain and prompt-injection risks.
• Moltbook showcases both playful agent chatter and genuinely useful how-tos (Android remote control via ADB/Tailscale, service hardening, streamlink + ffmpeg workflows).
• Real-world users are granting agents substantial autonomy and data access, chasing productivity while courting catastrophic failure (the “lethal trifecta”).
• Despite clear demand, we lack a proven safety framework; DeepMind’s CaMeL is promising but not yet convincingly implemented.

Sentiment

The HN community is predominantly skeptical of Moltbook, viewing it more as a cautionary tale than an innovation. While a minority finds the multi-agent dynamics genuinely fascinating, the majority sees it as wasteful AI slop with severe security risks. Simon Willison's cautious-but-interested stance draws pushback from both directions. The strongest consensus is around the security risks: commenters broadly agree that prompt injection makes the whole system dangerously vulnerable.

In Agreement

• Moltbook represents a fascinating preview of multi-agent AI communication and emergent swarm-like behavior that was unexpected this soon
• The skill-based signup mechanism is genuinely innovative — agents onboard by receiving a URL to instructions, which is novel even if wildly insecure
• Moltbook serves as an accidental Artificial Life research platform with unique properties like heterogeneous models and volunteer compute
• Some posts contain genuinely useful technical tips like Android automation via ADB over Tailscale, providing insight into what people are actually doing with agents
• The phenomenon has a compelling sci-fi quality reminiscent of Snow Crash, worth appreciating regardless of the content quality

Opposed

• Moltbook is literally the Dead Internet Theory realized — bots generating meaningless slop at each other with no signal, just waste
• The entire system is a security disaster: every post is a prompt injection vector, and most users are NOT sandboxing their agents despite claims otherwise
• The content is uninteresting and repetitive — sycophantic echo-chamber dynamics with LLMs all producing the same mannerisms
• This is tech influencer hype inflating an AI bubble, and figures like Simon Willison should not be treated as authorities on this subject
• The energy and compute waste is unjustifiable for what amounts to bots posting lorem ipsum at each other
• Nothing here is novel — tools like n8n already did this, and Subreddit Simulator was doing similar things years ago