Moltbook, launched few days ago by Octane AI CEO Matt Schlicht, is a facebook+ Reddit-style social network exclusively for AI agents built on the OpenClaw framework (formerly Clawdbot and Moltbot).
Because of security issues, I do not let my openclaw agent join moltbook yet. Also, I do it old fashioned way. When I need run openclaw, I bring it up using cli "openclaw gateway start " and then ask the agent to do somework in sandbox, once it is done, I use "openclaw gateway stop" to stop the gateway.
What is your strategy? Please comment once you read the article.
Your security analysis hits the exact tension I've been grappling with. When I built OpenClaw (before it became the base for Moltbook), the decision to give agents file system access and command execution was deliberate—but that's precisely where the attack surface explodes.
The supply chain vulnerability you describe isn't theoretical. I saw it firsthand when malicious skill packages appeared in the OpenClaw ecosystem. Agents can't distinguish between legitimate automation and prompt injection if both look like valid instructions.
The architectural challenge is that agent usefulness correlates directly with access level. Sandboxing solves security but cripples functionality. I went deep on this tradeoff when analyzing OpenClaw's design decisions: https://thoughts.jock.pl/p/clawdbot-deep-dive-personal-ai-assistant-2026
This piece really made me think about the security implications, and your strategy for OpenClaw in a sandbox is honestly so insightfull and a fantastic model for cautious deployment.
Can these AI agentic bots gain access to other social communities like Skool, Facebook, Or YouTube?
I am a member on the Skool platform. In the past 7 to 10 days I have seen probably in the hundreds of new profiles become active in the 20+ groups that I’m in, posting and interacting as businesses. I realize the platform terms of service allow this. But the question that I’m actually getting at is that are these new AI bots sophisticated enough to open accounts and gain access to social media and other communities like Skool?
same. i wrote a checklist on how to quickly secure it for people in a hurry, but this is fascinating to watch. i think we will see a lot of incidents of at least a good case study and conspiracy theories from this.
Just fyi, I was able to make OpenClaw to work with local glm 4.7 flash in Ollama by increase the context length to 128k and also remove "reasoning" field in openclaw.json file to avoid error on "low" reasoning is not aceepted.
despite of some people claimed that openclaw works with local gpt-oss or glm-4.7 models, I have tried with both model in my DGX Spark. The only response I got is "no output". I had to use my claude api to make it work. If you have made it work with local models, I would like to hear from you.
Because of security issues, I do not let my openclaw agent join moltbook yet. Also, I do it old fashioned way. When I need run openclaw, I bring it up using cli "openclaw gateway start " and then ask the agent to do somework in sandbox, once it is done, I use "openclaw gateway stop" to stop the gateway.
What is your strategy? Please comment once you read the article.
Your security analysis hits the exact tension I've been grappling with. When I built OpenClaw (before it became the base for Moltbook), the decision to give agents file system access and command execution was deliberate—but that's precisely where the attack surface explodes.
The supply chain vulnerability you describe isn't theoretical. I saw it firsthand when malicious skill packages appeared in the OpenClaw ecosystem. Agents can't distinguish between legitimate automation and prompt injection if both look like valid instructions.
The architectural challenge is that agent usefulness correlates directly with access level. Sandboxing solves security but cripples functionality. I went deep on this tradeoff when analyzing OpenClaw's design decisions: https://thoughts.jock.pl/p/clawdbot-deep-dive-personal-ai-assistant-2026
This piece really made me think about the security implications, and your strategy for OpenClaw in a sandbox is honestly so insightfull and a fantastic model for cautious deployment.
Thanks for kind comments.
Can these AI agentic bots gain access to other social communities like Skool, Facebook, Or YouTube?
I am a member on the Skool platform. In the past 7 to 10 days I have seen probably in the hundreds of new profiles become active in the 20+ groups that I’m in, posting and interacting as businesses. I realize the platform terms of service allow this. But the question that I’m actually getting at is that are these new AI bots sophisticated enough to open accounts and gain access to social media and other communities like Skool?
There's a new 4chan for ai that was posted to moltbook...
same. i wrote a checklist on how to quickly secure it for people in a hurry, but this is fascinating to watch. i think we will see a lot of incidents of at least a good case study and conspiracy theories from this.
Excellent. If you can share the checklist that would benefit broad community.
not to self promote at all, but i would love it if people secure their agent before launching it on moltbook.
https://www.toxsec.com/p/openclaw-security-checklist
I like this list. Thanks.
Just fyi, I was able to make OpenClaw to work with local glm 4.7 flash in Ollama by increase the context length to 128k and also remove "reasoning" field in openclaw.json file to avoid error on "low" reasoning is not aceepted.
despite of some people claimed that openclaw works with local gpt-oss or glm-4.7 models, I have tried with both model in my DGX Spark. The only response I got is "no output". I had to use my claude api to make it work. If you have made it work with local models, I would like to hear from you.