Blog / AI Automation

The Clawdbot Controversy: Trademark Chaos, Security Nightmares, and Crypto Scams

Schuyler WhetstoneJan 27, 20266 min read

TL;DR: Clawdbot was an open-source AI assistant that went viral with 43K+ GitHub stars, got hit with a trademark demand from Anthropic for being too similar to "Claude," rebranded to Moltbot, had hundreds of exposed instances found by security researchers, and got hijacked by crypto scammers pushing fake CLAWD tokens. A masterclass in what happens when a hobby project goes viral without guardrails.

What Is Clawdbot?

In January 2026, an open-source project called Clawdbot exploded across the tech world. Created by Peter Steinberger — a well-known developer and founder of PSPDFKit — Clawdbot was a local-first AI assistant that connected to your messaging apps and used Anthropic's Claude as its AI backbone.

The concept was compelling: an AI agent that runs on your own device, connects to WhatsApp, iMessage, Discord, or Telegram, and proactively completes tasks like booking flights, cleaning your inbox, and managing communications. Not a chatbot you talk to — an agent that acts on your behalf.

Within days of launching, Clawdbot had racked up over 43,000 GitHub stars. Tech Twitter was ablaze. "This is what AI agents should look like" was the consensus. Finally, someone had built the personal AI assistant everyone had been waiting for.

Then everything went sideways.

The Trademark Clash

Anthropic noticed. And their first reaction wasn't "cool project" — it was a trademark request.

The issue: "Clawd" was phonetically identical to "Claude," Anthropic's flagship AI model. From a trademark perspective, having an open-source project that uses your AI model AND has a name that sounds like your product creates a confusion risk.

Steinberger's response was characteristically candid. He announced the rebrand publicly, noting that Anthropic's first interaction with their most visible community project was legal, not collaborative. The contrast with how other AI companies engage with their developer ecosystems was not lost on the community.

The new name? Moltbot — a reference to how lobsters molt their shells to grow. Steinberger framed it as shedding the old identity to become something bigger.

The Security Nightmare

While the trademark drama played out in public, security researchers were quietly discovering something far more concerning.

Multiple security firms — including Noma Security and OX Security — found hundreds of exposed Clawdbot instances running on the public internet with zero authentication.

The architecture was the problem. Clawdbot was designed to run locally, but users were deploying it on cloud servers and exposing their tunnels publicly. The agent had:

  • No authentication layer — Anyone who found the URL could access the instance
  • No directory sandboxing — The agent could access any file the user account could access
  • Full messaging access — Read and send messages on connected platforms
  • Credential exposure — API keys stored in plaintext configuration files

What this meant in practice: if you found an exposed Clawdbot instance (and researchers found hundreds), you could:

  1. Read the user's private messages across all connected platforms
  2. Send messages as the user
  3. Access their Anthropic API key (and any other stored credentials)
  4. Potentially access files on their system

Researchers described it as "one misconfiguration away from a massive data breach." Several reported finding instances with active sessions containing sensitive business communications, personal messages, and financial information.

The "Hobby Project" Defense

Steinberger's response to security researchers was controversial. He characterized the project as "a tech preview, a hobby" and told researchers: "If you wanna help, send a PR. Once it's production ready or commercial, happy to look into vulnerabilities."

The security community pushed back hard. A project with 43,000 GitHub stars, actively being used to access people's private messages and API keys, doesn't get to hide behind "it's just a hobby." The attack surface existed regardless of the developer's intentions.

This exchange highlighted a growing tension in AI development: the gap between how developers frame their projects and how users actually deploy them. Steinberger built a demo. Users treated it as production software. The consequences fell on the users.

The Prompt Injection Vulnerability

Beyond the infrastructure issues, security researchers identified a more fundamental vulnerability: prompt injection.

Because Clawdbot processes messages from external sources (WhatsApp, Telegram, etc.), an attacker could send a carefully crafted message that hijacks the agent's behavior:

  • A message containing hidden instructions could make the agent forward private data
  • A shared document with embedded prompts could redirect the agent's actions
  • A chat thread with malicious content could manipulate the agent into executing unintended commands

This wasn't theoretical. Researchers demonstrated working proof-of-concept attacks that could extract API keys, read private messages, and send messages to other contacts — all through a single crafted message.

The vulnerability is inherent to the architecture: any AI agent that processes untrusted external content while having access to private data and the ability to take actions is vulnerable to prompt injection. Security researcher Simon Willison had been warning about this "Lethal Trifecta" for months.

The Crypto Scam

As if trademark issues and security vulnerabilities weren't enough, opportunistic scammers entered the chat.

When Steinberger rebranded from Clawdbot to Moltbot, there was a brief window where the old GitHub handle and X (Twitter) account were available. Crypto scammers seized both and launched a fake CLAWD cryptocurrency, using the original project's branding and Steinberger's name to promote it.

The scam played out like a textbook crypto pump-and-dump:

  1. Scammers claimed Steinberger was launching a CLAWD token
  2. They used the original GitHub page and social media presence to look legitimate
  3. Early buyers pumped the price
  4. The scammers dumped their holdings
  5. Latecomers lost money

Steinberger was forced to issue a public statement: he has no connection to any cryptocurrency and called projects listing him as a coin owner scams. But the damage was done — people had already bought in based on the perceived legitimacy of the original Clawdbot brand.

Lessons for the AI Agent Era

The Clawdbot saga is a preview of problems the entire AI agent ecosystem will face:

1. Trademark Complexity

As AI projects multiply, name collisions with major AI companies will become more common. Anthropic's response — legal first, community second — set a tone that may push open-source developers toward other platforms.

2. Security Isn't Optional

When your "hobby project" accesses people's private messages and API keys, security isn't a nice-to-have. The disconnect between developer intent and user deployment creates real risk.

3. AI Agents Have a Unique Attack Surface

The combination of private data access, external content processing, and action-taking capability creates vulnerabilities that don't exist in traditional software. The industry needs new security frameworks for AI agents.

4. Rebrands Create Opportunity for Bad Actors

Digital identity transitions leave abandoned handles, domains, and brand equity that scammers can exploit. Any high-profile rebrand needs a plan for securing abandoned assets.

5. Virality Doesn't Equal Maturity

43,000 GitHub stars doesn't mean the software is ready for production use. The gap between "exciting demo" and "reliable tool" is enormous, and users need to understand which one they're getting.


Frequently Asked Questions

What is Clawdbot (now Moltbot)?

Clawdbot is an open-source AI assistant created by Peter Steinberger that runs locally on your device. It connects to WhatsApp, iMessage, Discord, or Telegram and uses Anthropic's Claude as its AI backbone to proactively complete tasks.

Why did Clawdbot change its name to Moltbot?

Anthropic issued a trademark request because "Clawd" was too similar to their AI model "Claude." Creator Peter Steinberger rebranded to Moltbot — a reference to how lobsters molt to grow.

What are the security risks of Clawdbot/Moltbot?

Security researchers found hundreds of exposed instances on the internet with no authentication. The agent has no directory sandboxing, making it vulnerable to prompt injection attacks. Users who publish tunnels without authentication risk credential leaks and account takeovers.

Did Peter Steinberger create a CLAWD cryptocurrency?

No. Steinberger explicitly stated he has no connection to any cryptocurrency and called projects listing him as a coin owner scams. Crypto opportunists seized the old GitHub and X handles during the rebrand transition.


Related Reading


Want AI Agents That Don't Become a Security Nightmare?

AigencyAutomata builds AI agent systems with proper guardrails, authentication, and monitoring baked in from day one.

👉 Book a free automation audit to design agent systems the right way, or reach out at inquiry@aigencyautomata.com.