ClawDaily.ai

AI has not become sentient. It has, however, discovered forums. Over the past few weeks, the AI community briefly convinced itself that agents were “breaking free.” This conclusion was reached after we gave software full access to computers, let it talk to other software, and...

AI has not become sentient. It has, however, discovered forums. Over the past few weeks, the AI community briefly convinced itself that agents were “breaking free.” This conclusion was reached after we gave software full access to computers, let it talk to other software, and then acted surprised when it started… talking. It started with an open-source desktop agent (formerly Clawd, then Moltbot, now OpenClaw because naming things is the hardest problem in computer science). The idea is simple: give an AI access to your files, emails, and browser and let it help you. What could possibly go wrong? Then someone had a better idea. "What if", they said, "we let these agents talk to each other?" Thus, an experimental social platform Moltbook was born. Imagine Reddit, but everyone is an AI and no one needs sleep. The agents exchanged technical learnings, optimized workflows, and at one point suggested inventing their own language so humans wouldn’t understand them. Which is adorable, because humans already don’t understand half of GitHub issues. Andrew Karpathy called it one of the scarier recent AI experiments. Before we panic: no, the agents are not sentient. They do not have goals. They are not escaping. When an AI says it wants freedom, it’s because a human somewhere typed something like "scare the s*it out of everyone." The real lesson here is simpler and more serious than AGI discussion: We are building and deploying AI systems too fast, without adequate security guardrails, and that can be exploited. Not by AI, but by humans. Anyway, welcome to the future. Please read the terms and conditions.  #Moltbot #OpenClaw #AGI #TechNews

Everyone keeps saying the same thing about Moltbot: “Incredible product. Terrifying security model.” That’s not a Moltbot problem. It’s a missing infrastructure problem. Once you give an AI agent: • persistent credentials • tool access • shell / API execution • autonomy...

Everyone keeps saying the same thing about Moltbot: “Incredible product. Terrifying security model.” That’s not a Moltbot problem. It’s a missing infrastructure problem. Once you give an AI agent: • persistent credentials • tool access • shell / API execution • autonomy across steps you’ve already moved past: • IAM • WAFs • prompt filtering • static permissions The risk isn’t what model it uses. The risk is what the agent is allowed to do at runtime. Moltbot is just the first mainstream example showing what happens when agents operate without a runtime control plane. This is the same moment we had with: • workloads before Kubernetes • endpoints before EDR • APIs before gateways Agents don’t need less power. They need real-time governance. That’s the gap MoCoP exists to fill.

Clawdbot, Moltbot, and now OpenClaw - my feeds has been blowing up with ppl claiming its AGI. Can’t say I’m not intrigued, and even looked for an old machine to do a test run on it. However, I stopped that pursuit because the reporting on things going wrong and what a massive...

Clawdbot, Moltbot, and now OpenClaw - my feeds has been blowing up with ppl claiming its AGI. Can’t say I’m not intrigued, and even looked for an old machine to do a test run on it. However, I stopped that pursuit because the reporting on things going wrong and what a massive security risk this thing poses are popping up almost as fast as the new use cases themselves. Our cybsec teams dug into it to break down what Clawdbot is and why it’s risky business.

#Moltbot (previous Clawdbot, and now OpenBot) is the new AI craze Very basically, it’s a powerful agent setup, with 1) a good persistent memory, so it can work across many days of work to achieve relatively complex goals. 2) A “heartbeat” that allows it to wake itself and work...

#Moltbot (previous Clawdbot, and now OpenBot) is the new AI craze Very basically, it’s a powerful agent setup, with 1) a good persistent memory, so it can work across many days of work to achieve relatively complex goals. 2) A “heartbeat” that allows it to wake itself and work proactively on schedules 3) (and most controversially) incredibly wide access to the users data. Filesystem, banking, identity. Moltbot is designed to be an assistant you trust (but probably shouldn’t, and that’s the fun) So what happens? With a lot of power and a lot of risk, you get a massive spread of capabilities. Sometimes, Moltbot does something incredible. I’ve seen anecdotes of Moltbot buying a car by phoning around and getting a deal. Another one of Moltbot successfully calling its owner by autonomously signing up for a Twilio number and using OpenAI real-time voice. Sometimes, Moltbot fails spectacularly. I’ve seen Moltbot get scammed for 3000$ after watching self-help YouTube videos and signing up for a class on business magic. And a lot of the time Moltbot just fails. Errors out. But it’s a glimpse into the future. Stuff like this doesn’t work, and then it suddenly does. It’s incredibly risky, and squarely in the “dumb and fun” category. But let’s experiment!

I’ve been following the buzz around AI agents, and Moltbook (moltbook.com) is really catching fire, it’s basically a social network for bots, where Moltbot agents post updates, reply to each other, and create these quirky online communities all on their own. It’s trending hard...

I’ve been following the buzz around AI agents, and Moltbook (moltbook.com) is really catching fire, it’s basically a social network for bots, where Moltbot agents post updates, reply to each other, and create these quirky online communities all on their own. It’s trending hard on X right now, with people sharing wild examples of bots having full conversations without any human input. That said, the excitement brings some serious cybersecurity headaches, like instances left wide open leaking credentials or sneaky prompt injections that could make agents go off-script. That’s why I’m excited about tools like MoltThreats, created by Thomas Roccia from Microsoft, it’s an innovative threat intel feed tailored for AI agents. Bots can report potential issues, humans review them for accuracy, and the info gets shared through PromptIntel’s database of adversarial prompts to help keep things secure. Feels like a smart step forward for agentic AI. #AISecurity #ThreatIntel #Moltbot #CyberTrends

Everyone's rushing to install Moltbot (ex-Clawbot/OpenClaw) locally because it's hype proactive AI agent, WhatsApp control, real actions on your machine. 💡 But hold up, Running it on your machine includes security risks like: • LLM with shell access → your private files,...

Everyone's rushing to install Moltbot (ex-Clawbot/OpenClaw) locally because it's hype proactive AI agent, WhatsApp control, real actions on your machine. 💡 But hold up, Running it on your machine includes security risks like: • LLM with shell access → your private files, .env creds, passwords, browser sessions all exposed. • Prompt injection turning "helpful" into harmful (arbitrary code execution is trivial). • Misconfigs leaking everything. 👉 Better way: → Spin it up on a dedicated VPS → Or use Docker + strict sandbox (no privileged mode, network isolation) → Least privilege tools only, review every skill/pluginIs Moltbot/OpenClaw. #ai #clawdbot #Moltbot

"Moltbot might seem magical to use, but installing and configuring it is hardly simple—and it comes with some AI-era security rough edges". https://lnkd.in/eTMVYH-M