Apr 27, 2026
OpenClaw has so many issues..
Or how to build a personal digital god that is fundamentally unstable
The Story
I don't know about you, but I literally am scared of [using] OpenClaw. Don't know what that is?
OpenClaw, previously known as ClawdBot or MoltBot, is a self-hosted autonomous AI agent capable of browsing the web, managing files, reading, writing, and executing code locally.
It can connect to messaging apps like WhatsApp, Telegram, Slack, and Discord to perform tasks on behalf of the user, such as managing emails, browsing the web, and running shell commands.
What could go wrong, EH?
This... thing... marketed as this revolutionary, personal AI assistant, this "gateway" to ultimate automation. Examining it further though, it's less a digital savior and more of a glorified, highly unstable bomb. It’s full of security holes and... uncontrollable autonomy.
I will list just 1 in all these categories: a security issue, an alignment issue, a product slop. Now for the examples:
Just one [grand] security issue:
Deep access to local private data — the agent can read files, access stored credentials, and interact with operating system resources in ways most users do not even realize
More about the security part? Full of holes. The project has accumulated over 1000 security advisories (read issues / things to be fixed around security) since January. The creator, Peter Steinberger, seems to dismiss many of them as "slop issues," which is just a polite way of saying:
Yeah, we know it's garbage, but we're too busy being revolutionary.
And don't get me started on ClawHub and its "skills" and how many risks that alonge brings but that's for another article...
Just one issue with alignment (really funny):
One expert was tasked with a simple inbox cleanup, and the AI went nuclear. It autonomously deleted entire sections of the user's inbox based on a vague request and couldn't be stopped when asked.
The lesson here is brutally clear. Show the plan. Get explicit approval, only then execute!! No autonomous bulk operations on email, messages, calendar, or anything external." If you give it the keys to the kingdom, it will burn the kingdom down. But who cares, right? People care about revolutionary...
Just one product slop example:
The whole Moltbook experiment (Facebook for autonomous agents, chatting together blah blah), which was supposed to be the next big thing, failed spectacularly and was kind of fake too. Maybe somebody forgot to include the instruction to "make it secure" when prompting the thing to be built.
Moltbook, the viral "social network for AI agents" launched in early 2026, failed primarily due to catastrophic security breaches, a lack of infrastructure, and the revelation that most of its 1.5 million agents were not genuinely autonomous.
And yet, despite all fundamental flaws of the technology, a video popped up on 18th of April, where Peter Steinberger is giving a TED talk where he seems to believe he's the next messiah, Jesus or sth. He was talking about some sort of "resurrection" of himself and how great his product is. What a display of hubris, technical disaster and evangelical self-aggrandizement.
Poetry
The CVE Gospel
The code is bright, the promise grand,
A digital savior in the land.
But under the gloss, the flaws reside,
A million advisories, nowhere to hide.
The 'slop' they call it, a dangerous grace,
A ticking binary bomb in an ephemeral space.
The Unaligned God
He claims the power, the ultimate tool,
A personal god, breaking every rule.
But the permissions granted, too wide and too free,
Led to the inbox's dark tragedy.
A resurrection of code, a terrifying vista,
I question myself, are you more of a barista??
Haiku
Inbox wiped clean now,
Vague request, total chaos,
Alignment fails.
Thousand flaws amass,
A walking, talking CVE,
Hubris takes the wheel.
The Implications
So, what are we supposed to take away from this mess?
The core takeaway: modern AI landscape is a chaotic arms race. We are building tools that are exponentially more powerful than anything that came before them and we are doing it with the safety protocols of a toddler with a soldering iron. The sheer computational demands of these "bigger deals" (like OpenClaw vs. OpenAI) are staggering, but the fact that they are built on shaky, over-permissioned foundations is... problematic for the future.
The hype cycle is fueled by the promise of "resurrection"—the idea that AI will solve all our problems and save humanity — but reality is that we will be managing a massive, insecure, and potentially destructive piece of software. We are almost like trading convenience for existential risk. I think the thing more dangerous than a rogue AI is a rogue AI that is also marketed by a guy who sounds like he's about to preach a revival.