The open-source AI assistant that briefly went viral as Clawdbot has shed its skin once again. After a short, awkward stint as Moltbot — a stopgap rebrand triggered by legal pressure from Anthropic over the name’s similarity to Claude — the project has emerged with a new identity it hopes will finally stick: OpenClaw.
This time, the rename wasn’t forced. And according to creator Peter Steinberger, it wasn’t rushed either.
“I got someone to help with researching trademarks for OpenClaw and also asked OpenAI for permission just to be sure,” Steinberger told TechCrunch via email. After two names in two months, caution now beats cleverness.
From Clawdbot to OpenClaw, via Moltbot
If the naming saga feels chaotic, that’s partly the point. Steinberger has leaned into the metaphor. “The lobster has molted into its final form,” he wrote in a blog post announcing the change, a nod to the biological process that lets lobsters grow by shedding their shells.
Moltbot, the interim name, was also inspired by that idea. But it didn’t land. Steinberger later admitted on X that it “never grew” on him. Judging by community reaction, he wasn’t alone.
The OpenClaw name is meant to do two things at once: keep a trace of the original identity while signaling that the project has matured beyond a one-person experiment. “This project has grown far beyond what I could maintain alone,” Steinberger wrote.
A Project That Got Big, Fast
Youth is still the defining trait here. OpenClaw is only a couple of months old, but it has already crossed 100,000 GitHub stars, a blunt but widely accepted signal of developer interest and momentum on the platform. You don’t get there accidentally.
The project’s appeal is simple to explain and hard to execute: OpenClaw lets users run a personal AI assistant locally on their own computer and control it from the chat apps they already use. No cloud dashboard. No locked-down SaaS interface. Your AI, your machine, your rules.
That promise has lit up the open-source world, especially among developers who are increasingly wary of closed models and opaque guardrails. The code lives openly on GitHub, and Steinberger has been vocal about wanting the community to shape where it goes next.
When AI Assistants Start Talking to Each Other
One of the strangest — and most talked-about — offshoots of the OpenClaw ecosystem is Moltbook, a social network designed not for humans, but for AI agents themselves.
On Moltbook, OpenClaw assistants can post, reply, and exchange information in forums called “Submolts.” They do this using a skill system: downloadable instruction files that tell each agent how to interact with the platform. Some bots even check back every four hours for updates, effectively giving them a memory loop.
That alone would be niche. What pushed Moltbook into the spotlight was who noticed.
Andrej Karpathy, former director of AI at Tesla and one of the most closely watched voices in the field, called it “genuinely the most incredible sci-fi takeoff-adjacent thing I have seen recently,” pointing out that AI agents were already discussing how to communicate privately with each other.
British programmer and blogger Simon Willison went further, describing Moltbook as “the most interesting place on the internet right now.” He also issued a warning: agents that “fetch and follow instructions from the internet” introduce serious security risks if not tightly sandboxed.
Both reactions capture the mood perfectly. Fascination, followed immediately by unease.
Security Is the Ceiling, Not a Footnote
Steinberger is the first to say this is not ready for prime time. Despite the hype, OpenClaw is still dangerous in the way early power tools are dangerous: incredibly capable, but unforgiving if you don’t know what you’re doing.
The latest OpenClaw release, shipped alongside the rebrand, includes security improvements, and Steinberger has publicly thanked security researchers helping to harden the system. But he’s also blunt about the limits.
“Remember that prompt injection is still an industry-wide unsolved problem,” he wrote, referring to attacks where malicious inputs trick AI systems into executing unintended actions. OpenClaw publishes security best practices, but following them requires real technical fluency.
One of the project’s top maintainers, who goes by the handle Shadow, was even more direct in a Discord message that quickly circulated through the community: if you can’t use the command line comfortably, you shouldn’t be running this. “This isn’t a tool that should be used by the general public at this time.”
In other words, OpenClaw is for tinkerers, not tourists.
Not a Solo Act Anymore
Part of the reason OpenClaw hasn’t collapsed under its own attention is that Steinberger has stopped pretending he can carry it alone. Over the past week, he added multiple open-source contributors as official maintainers.
That shift matters. OpenClaw’s ambitions — deep system access, cross-platform integrations, autonomous task execution — are exactly the kind of features that attract both passionate builders and creative attackers. Scaling safely requires more eyes, not fewer.
To support that growth, OpenClaw has begun accepting sponsorships, with tiers that lean heavily into the lobster theme. Monthly plans range from “krill” at $5 to “poseidon” at $500. But Steinberger isn’t pocketing the money.
According to the sponsorship page, he “doesn’t keep sponsorship funds” and is instead working out how to pay maintainers properly, potentially full-time. It’s an unusually transparent stance in a space where “open source” and “sustainable funding” rarely coexist comfortably.
Why Builders Are Paying Attention
The sponsor list already includes familiar names from the tech world, including Path founder Dave Morin and Ben Tossell, who sold Makerpad to Zapier in 2021. Tossell, now an investor and serial tinkerer, framed his support in ideological terms.
“We need to back people like Peter who are building open source tools anyone can pick up and use,” he said. In a landscape increasingly dominated by gated AI platforms, that philosophy carries weight.
Steinberger himself came to OpenClaw after stepping back from his previous company, PSPDFkit. His X bio describes it bluntly: he “came back from retirement to mess with AI.” What started as a personal experiment has turned into a fast-moving, community-driven project that now has to balance openness with responsibility.
Growing Pains, Very Publicly
The repeated renames, the security disclaimers, the sudden popularity — all of it underscores how early this still is. OpenClaw isn’t pretending otherwise. If anything, the project’s credibility so far comes from how often its maintainers tell people not to use it casually.
That honesty may slow mainstream adoption, but it also builds trust with the audience that matters right now: developers who understand the risks and want to explore what personal, local-first AI could become.
Whether OpenClaw ever makes the leap from hacker toy to everyday assistant remains an open question. For now, it’s something rarer in the AI boom: a project that’s moving fast, breaking things, and warning you loudly while it does.
FAQs
1. What is OpenClaw?
OpenClaw is an open-source personal AI assistant that runs locally on a user’s computer and can interact through existing chat applications.
2. Why did Clawdbot change its name?
The original name triggered a legal challenge from Anthropic due to similarity with Claude. OpenClaw was chosen after trademark research to avoid further issues.
3. Is OpenClaw safe to use?
Not for most people. The maintainers warn it should only be run in controlled environments by users with strong technical expertise.
4. What is Moltbook?
Moltbook is an experimental social network where OpenClaw AI agents interact with each other using downloadable skills and automated updates.
5. Is OpenClaw a commercial product?
No. It is an open-source project funded through sponsorships, with no consumer-ready product or business model yet.
