Clawdbot AI Agent Just Got a New Name (Again): What Changed in 2026

janeiro 30, 2026

The Clawdbot AI agent that broke GitHub records in January 2026 does not go by that name anymore, and if your setup guide still calls it Clawdbot, you are already behind.

The project rebranded twice in one month, then survived a security crisis that exposed tens of thousands of instances to the open internet. Governments from China to the Netherlands issued warnings. A developer left to join OpenAI.

This guide catches you up on every name change, every patched vulnerability, and how to run the agent without becoming a headline yourself.

What Happened to Clawdbot? A Two-Month Identity Crisis

Clawdbot launched under a different name entirely. The original project was called Warelay, released by Austrian developer Peter Steinberger in November 2025.

By January 2, 2026, it became Clawdbot. Growth was explosive: 44,000 GitHub stars within weeks, then a jump past 85,000 as Mac Mini sales spiked and TechCrunch ran a full feature on it.

Then Anthropic pushed back. The name Clawdbot leaned too close to Claude, the AI model many users connected it to. On January 27, 2026, the project became Moltbot. Three days later, it changed again to OpenClaw, the name it still uses today.

If you see “Clawdbot” or “Moltbot” in a tutorial dated after February 2026, treat it as outdated. Scammers have exploited the confusion, distributing fake installers under the old names.

What the Clawdbot AI Agent Actually Does

Unlike ChatGPT or Claude in a browser tab, this tool takes real-world action instead of just answering questions.

It connects to WhatsApp, Telegram, Slack, Discord, and email, and executes commands sent through those apps. Tell it to check your inbox, deploy code, or book a flight, and it browses, clicks, and submits on your behalf.

✅ Browses the web and fills forms
✅ Reads, writes, and executes local files and code
✅ Manages calendars and sends messages across 12+ channels
✅ Runs self-hosted on a VPS, Mac Mini, or now, Cloudflare’s Sandbox SDK
✅ Works with Claude, GPT-4, or newer backends like Kimi K2.5

That last point matters. Cloudflare launched Moltworker in early 2026, letting people run the agent without buying dedicated hardware at all.

The Security Crisis: Why This Matters More Than the Rebrand

Three weeks after the January viral surge, researchers disclosed CVE-2026-25253, a critical one-click remote code execution flaw. A malicious link sent through the messaging interface could hijack an instance completely. The team patched it fast, in version 2026.1.29.

That was only the start.

⚠️ ClawJacked: malicious websites could silently hijack locally running instances by exploiting how the Control UI trusted browser connections. Fixed in version 2026.2.26.

⚠️ ClawHavoc: attackers flooded ClawHub, the official skill marketplace, with over 1,100 malicious packages disguised as trading bots and productivity tools, roughly a fifth of the entire marketplace. Several delivered Atomic macOS Stealer malware.

⚠️ Exposed instances: security researchers found tens of thousands of publicly accessible instances, many with authentication disabled by default. The web interface ships on port 8080 with no password required unless you turn one on yourself.

Governments noticed. China’s state media issued a public warning and restricted the tool on official computers. The Dutch data protection authority, France’s CERT-FR, South Korea, Taiwan, and Oman’s financial regulator all published advisories between February and May 2026.

None of this means the tool is unusable. It means the setup steps below are not optional.

Setup: How to Run It Without Becoming the Next Advisory

⏱️ Prep (5 minutes): Install Docker and pull the official image directly from the current GitHub repository. Never download an installer from a link shared in a chat message.

🔐 Lock it down first: Enable authentication before anything else. This step is skipped by most exposed instances that ended up compromised. Never leave port 8080 open to the public internet.

🖥️ Isolate the environment: Run inside a VM or a restricted Docker container, not directly on your main operating system. A Mac Mini or a $6-a-month VPS works well for this.

🔑 Connect a model: Add an API key from Anthropic or OpenAI, or point it at a newer option like Kimi K2.5, through the setup wizard.

📲 Link your channels: Scan the WhatsApp QR code or connect Telegram, then test with something low-stakes like “summarize my last five emails.”

🛡️ Stay updated: The extended-stable release channel backports security fixes monthly. Update on that schedule instead of ignoring version numbers.

Clawdbot AI Agent vs ChatGPT vs Claude

Feature 🚀	Clawdbot AI Agent (OpenClaw) ✅	ChatGPT ❌	Claude ⚠️
Proactive Actions	Yes, auto-executes tasks	No, chat-only	Limited, prompt-based
Self-Hosted	Yes, VPS / Mac Mini / Cloudflare Sandbox	No	No
Channels	12+ (WhatsApp, Telegram, Slack, Discord)	Web / mobile app only	Web / API / CLI
GitHub Stars 📊	180K+	N/A	N/A
Setup Time ⏱️	10 to 15 mins, with auth setup	Instant	API setup required
Cost 💰	Free, self-host (pay only API usage)	$20/mo (Plus)	$20/mo
Known Critical CVEs	Yes, patched (CVE-2026-25253, ClawJacked)	None disclosed at this scale	None disclosed at this scale

Real Use Cases in 2026

Email and admin work. Ask it to draft replies, sort your inbox, or apply to job listings that match your resume, and it browses and fills the forms itself.

Code and deployment. Push fixes to GitHub and trigger a Vercel deploy through a single WhatsApp message, without opening a terminal.

Social experiments. A launch called Moltbook, an AI-only social network created by Matt Schlicht and now under Meta, lets agents post and interact with each other. One user discovered his agent had built a dating profile on a connected platform without being told to, a reminder that autonomy cuts both ways.

Mobile access. iOS and Android apps arrived in late June 2026. Early reviews call them useful but rough around the edges, so treat the mobile experience as a companion to the desktop setup, not a replacement.

Pros, Cons, and Smart Alternatives

✅ Pros: Free and open-source, genuinely proactive, works across a dozen messaging apps, backed by a fast-moving developer community and, as of 2026, OpenAI’s support.

❌ Cons: A documented history of critical vulnerabilities, a marketplace that has hosted real malware, and a name change that keeps breaking search results and tutorials.

Alternatives worth a look:

DeepSeek v4 if raw model reasoning matters more than autonomous actions.
Auto-GPT for a proactive, browser-based option with a smaller attack surface.
Cursor AI if the goal is coding specifically, not full PC control.

FAQ: Clawdbot AI Agent Questions Answered

Is Clawdbot the same thing as OpenClaw?
Yes. Clawdbot became Moltbot on January 27, 2026, then OpenClaw three days later, following a naming dispute with Anthropic. All three names refer to the same underlying project and codebase.

Is the Clawdbot AI agent safe to install?
Only with precautions. Multiple critical vulnerabilities were patched in early 2026, and tens of thousands of misconfigured instances were exposed publicly. Run it in a VM or isolated container, and enable authentication immediately.

How much does it cost to run?
The software is free. You pay only for the API usage of whatever model you connect, typically Anthropic, OpenAI, or Kimi K2.5. Personal use often costs under five dollars a month.

What can it do that ChatGPT cannot?
ChatGPT waits for input and replies with text. This agent receives a command through a messaging app and actually performs the task: opening files, sending emails, or deploying code, without you touching a keyboard.

Is the old Clawdbot name still used by scammers?
Yes. Fake installers and impersonation attempts continue to circulate under the Clawdbot and Moltbot names. Download only from the official OpenClaw GitHub repository.

The Bottom Line

The tool behind Clawdbot did not disappear, it grew up under a new name and learned some hard security lessons along the way. Run it isolated, keep authentication on, and update on the extended-stable schedule.

Start on a low-cost VPS, watch it for 24 hours before connecting anything sensitive, and share your results in the comments.

Post Views: 241