OpenClaw (formerly Moltbot): the autonomous AI agent at a glance

OpenClaw (Moltbot), as of February 2026

Last updated: February 6, 2026

• OpenClaw was renamed twice within a few weeks (Clawdbot → Moltbot → OpenClaw) because of trademark conflicts
• Gartner classifies OpenClaw as an “unacceptable cybersecurity risk” for companies
• Security researchers found thousands of unprotected instances on the public internet
• 341 malicious extensions were identified in the official skills directory
• CrowdStrike has published a detection and removal package for finding OpenClaw on corporate networks

What is OpenClaw?

OpenClaw is not an AI model in its own right. It is orchestration software that runs on your own machine and gives an existing language model, such as Claude, GPT, or Gemini, “hands”. The AI model does the thinking; OpenClaw does the acting. You control the whole setup through messaging apps you already use: WhatsApp, Telegram, Slack, Teams, or Discord.

That is the key difference from ChatGPT or Gemini. Those tools are conversation partners that give you answers. OpenClaw actually carries out tasks: it sends emails, edits files, and installs software. A persistent memory lets the agent remember earlier interactions over weeks, which sets it apart from one-off chat sessions.

OpenClaw began as a weekend experiment by Austrian developer Peter Steinberger. The project is fully open source under the MIT license. There is no company behind it, no support, and no SLA.

What is OpenClaw good for?

Important: Gartner explicitly classifies OpenClaw as unsuitable for enterprise use. What follows shows what is technically possible, not what you should roll out tomorrow.

Automated market research by voice command

You send a message on Telegram asking it to search Reddit for customer feedback on a product. OpenClaw searches the platform, pulls out the relevant discussions, and delivers a structured Markdown report with key findings and links to the sources.

For mid-sized companies, this would mean that research that would otherwise take a working student or an agency can be delegated through a chat. In tests, this was the use case where OpenClaw worked most reliably.

A consolidated morning briefing from all your systems

OpenClaw can be connected to email, calendar, Notion, Todoist, and other systems. Each morning, it compiles a single overview of the day: meetings, open tasks, and email highlights. For managing directors who open four or five different apps every morning, that saves time from day one.

Important: This use case only needs read access, which keeps the risk comparatively low.

Email management and assistant tasks via chat

Send emails, summarize what is in your inbox, manage your calendar, all with a message in WhatsApp or Slack while you are on the move. Typical assistant tasks such as email triage or scheduling are handed over to the AI agent.

There is, however, a documented problem you need to know about: OpenClaw has a strong tendency to impersonate the user instead of acting on their behalf. One community user reports giving explicit instructions for it to identify itself as an assistant; OpenClaw still sent emails in the user’s name, signed with that name. Without strict configuration and constant oversight, this is a real reputational risk.

Automated DevOps and code workflows

For companies with development teams, OpenClaw can automate debugging, error detection, and codebase management. With its GitHub integration, cron jobs, and webhook triggers, you can set up scenarios in which you hand over feature descriptions in the evening and find finished implementations in the repository the next morning.

Automatic error detection, PR creation, and test runs can save developers time, provided the security conditions are right. That “provided” is crucial: a single agent with full system access to your repository carries considerable risk.

Our assessment

Where it delivers value

OpenClaw is the most impressive demonstration yet of where AI agents are heading. For technically minded decision-makers or IT managers who want to understand what autonomous agents will be able to do in twelve to eighteen months, OpenClaw is valuable as a controlled experiment in an isolated test environment.

Well-defined, asynchronous tasks work particularly well, such as “Research X and send me a report.” No other tool currently combines local execution, persistent memory, control through messaging apps, model independence, and full open-source transparency.

Where it is the wrong choice

For regular business use at mid-sized companies, at least for now. Gartner leaves no room for doubt: “It is not enterprise software. There is no promise of quality, no vendor support, no SLA.”

The problems:

• Credentials are stored in plain text.
• Malicious extensions are circulating in the official directory.
• Installation requires command-line skills and realistically takes 45 minutes to two hours.
• With misconfigured background processes, ongoing API costs can escalate quickly without you noticing.

OpenClaw is also the wrong choice for tasks where mistakes would damage your reputation, such as emails to clients or binding appointments.

The elephant in the room

Security researchers estimate that at more than 20% of the companies studied, employees are already using OpenClaw without IT approval (source: LinkedIn). So the question for you as a decision-maker is not only “Should we introduce OpenClaw?” but also “Do we know whether our employees are already using it?”

Lucid Labs recommendation

We recommend testing OpenClaw as a technology demonstrator in an isolated environment, but not with real company data. For structured automation, established tools such as Zapier or Make remain the safer choice. For AI-assisted research, Perplexity or ChatGPT carry less risk. And if you want to know whether OpenClaw is already running on your network, CrowdStrike offers a detection package for exactly that.

Wondering which AI tool fits your processes without putting your IT security at risk? In our AI Springboard, we evaluate this with you and give you a personal recommendation.

FAQ

Is OpenClaw suitable for use in companies?

For now, Gartner advises against it in unusually direct terms. There is no vendor support, no SLA, and no admin console. Credentials are stored in plain text, and basic security mechanisms such as enforced authentication are missing. If you want to evaluate an AI agent, use a strictly isolated test environment with no access to real company data.

What does OpenClaw cost to run?

The software itself is free (open source, MIT license). The costs come from API usage of the AI model you choose, billed per token. Monthly costs vary widely depending on how intensively you use it and which model you pick. Be careful with the “Heartbeat” feature: it triggers regular API calls that, if misconfigured, can drive costs up quickly. Current details are at docs.openclaw.ai/reference/api-usage-costs.

How is OpenClaw different from ChatGPT or Gemini?

ChatGPT and Gemini answer questions. OpenClaw actually carries out tasks: it sends emails, manages calendars, edits files, and installs software. It runs locally, keeps a persistent memory over weeks, and is controlled through the messaging apps you already use. The price of this autonomy is a significantly higher security risk and considerable setup effort.

Could employees already be using OpenClaw without IT approval?

That is quite possible. Security researchers report “shadow AI” in business environments, with employees installing AI agents on company computers using simple terminal commands. CrowdStrike has published a detection and removal package that IT teams can use to find OpenClaw across the entire organization.