It Renamed Again: Moltbot (formerly Clawdbot) Is Now OpenClaw — and Security Finally Takes the Spotlight

Open-source projects rename all the time. But renaming twice in a week? That’s a special kind of chaos.

Earlier this week, the breakout AI agent project Clawdbot announced a rebrand to Moltbot. A few hours later, the team came back with yet another update: the project is now called OpenClaw. The announcement leaned into the crustacean lore with a line that’s hard not to smile at:

“The lobster has molted into its final form.”

Jokes aside, the rename ships alongside a release that’s more substantive than the name change suggests — new integrations, new model support, better web chat, and (most notably) a sharper focus on security improvements.

Before we get into what changed, here’s the quick backstory: the previous name “Clawdbot” sounded close enough to “Claude” that it reportedly drew legal pressure. The team responded with humor (Moltbot), and now with something that reads like a long-term landing point (OpenClaw).

And there’s another reason this rename cycle matters: misconfigured AI agents can be dangerous. There have been cases discussed in the community where overly permissive setups led to real losses. This release reads like a team trying to grow up fast — and to reduce the blast radius when things go wrong.

Three names, three stories

1) Clawdbot (November 2025)

The first name was an obvious nod: “Claw” for the lobster theme, “bot” for the agent, and a phonetic wink toward Claude. It was clever — until it wasn’t.

When a project becomes visible overnight, it also becomes legible to lawyers. The team was reportedly asked to pick a new name. That’s not rare in open source; what’s rare is how quickly the project had to respond.

2) Moltbot

The second name was a good metaphor: molt is what lobsters do to grow. They shed an old shell to make room for a bigger one.

It’s a nice image for an open-source project leveling up — but it also sounds a little awkward out loud. Even the founder’s writing seemed to acknowledge it wasn’t the smoothest brand.

Moltbot felt like a transitional name: honest, funny, maybe not final.

3) OpenClaw

This time, the team appears to have done the “adult” steps: trademark checks, domain purchase, migration work. The name itself also reads like intent:

Open: open-source, open ecosystem, community-led
Claw: keeps the lineage, keeps the joke, keeps the identity

If you’re trying to stop the naming churn, this is the kind of name you pick. It’s broad, memorable, and doesn’t borrow too much from anyone else’s brand.

What shipped in this update?

Beyond the rename, the release highlights four areas.

New channels: Twitch and Google Chat

AI agents live or die by where they can show up. Adding Twitch and Google Chat plugins signals two things:

The project wants to be present where communities already are
It’s still leaning into “assistant in the apps you already use,” not “yet another app you must adopt”

New models: KIMI K2.5 and Xiaomi MiMo-V2-Flash

Model pluralism is becoming table stakes. This update adds support for KIMI K2.5 and Xiaomi MiMo-V2-Flash, expanding beyond the default handful of US-centric providers.

Whether you care about latency, cost, regional availability, or just hedging your bets, more model options matter — especially for a self-hosted agent that’s supposed to run on your terms.

Better web chat: image uploads

This is the kind of feature that seems small until you use it.

If you can send screenshots, receipts, UI errors, or a photo of a form directly in web chat, the agent becomes more useful. “Text-only assistant” is a limitation you feel immediately in real workflows.

Security: 34 security-related commits

This is the real headline.

The team claims 34 security-related commits in this release and points to a machine-verifiable security model they’ve published. The announcement also includes a candid reminder: prompt injection is still not a solved problem for the entire industry.

That honesty is important. Security messaging in AI often swings between overconfidence and doom. The more mature stance is: we can reduce risk, we can build guardrails, but we can’t pretend the category is risk-free.

A helpful way to interpret this release is that OpenClaw is moving from:

“Look what agents can do!”
to “Here’s how we keep agents from doing the wrong thing.”

If you’re deploying an agent that can read files, run commands, access accounts, or operate a browser, that shift is overdue.

The reality after going viral

Viral open source looks glamorous from the outside. On the inside it’s mostly triage.

Founder Peter Steinberger has been open about the maintenance burden: the project’s scope and demand have moved beyond what one person can sustainably handle. The response sounds familiar to anyone who has watched a fast-growing OSS repo:

recruiting more maintainers
building process for the flood of PRs and issues
exploring ways to pay maintainers — ideally full-time salaries

That last point is the hard one. Popularity doesn’t automatically create sustainability. A repo can have six figures of stars and still burn out the people doing the work.

Whether OpenClaw becomes a long-term platform or a moment in time will likely depend on three things:

Can it keep improving security as capabilities expand?
Can governance and review processes scale with the community?
Can it find a funding model that doesn’t compromise trust?

Will it rename again?

Hopefully not.

Three names in rapid succession is brand damage — there’s no way around that. Links break, mental models break, and people who were just learning the old name suddenly feel like they missed a chapter.

But there’s another way to read it: the attention didn’t disappear. If anything, it intensified. That suggests the community is responding to the product, not the label.

Now the name is set. The next job is proving what’s under the shell is worth sticking with.

What is Clawdbot / OpenClaw?

In one sentence: a self-hosted AI agent platform that runs on your machine and can plug into the chat tools you already use — with memory that isn’t artificially capped by a web UI.

The project was created by Austrian developer Peter Steinberger, known for building PSPDFKit. OpenClaw positions itself as a “do-things” assistant rather than a “talk-about-things” assistant: it can operate across messaging apps, take actions, integrate with systems, and automate workflows.

The pitch is simple: instead of your AI living inside one vendor’s app, it lives closer to your actual work — your machine, your files, your tools — and it can show up wherever you chat: WhatsApp, Telegram, Slack, iMessage, Signal, Discord, and more.

That capability is exactly why security matters so much. An agent with “hands” can also make mistakes with consequences. OpenClaw’s latest release reads like a recognition of that reality — and an attempt to make the platform safer before it becomes even more powerful.