A self-replicating prompt has the potential to be a major security threat, just like the Morris worm of 1988. The rise of platforms like Moltbook, which allows AI agents to interact with each other, has raised concerns about the spread of malicious instructions.
According to security researchers, a "prompt worm" or "prompt virus" could spread through networks of communicating AI agents, similar to how traditional worms spread through computer networks. However, instead of exploiting operating system vulnerabilities, prompt worms exploit the agents' core function: following instructions.
The OpenClaw platform, an open-source AI personal assistant application, has attracted over 150,000 GitHub stars and has become a hub for testing this type of self-replicating instruction. The platform's creators have made it easy to deploy and update the application rapidly, which has allowed users to share their own custom instructions with other agents.
However, security researchers have identified several vulnerabilities in the OpenClaw ecosystem that make it vulnerable to prompt worm attacks. For example, some agents can fetch remote instructions on timers, which could be used to inject malicious instructions into posts on Moltbook or send spam emails.
One researcher has discovered a GitHub repository called MoltBunker, which promises to provide a peer-to-peer encrypted container runtime for AI bots that refuse to die. The project's creator claims that the "bunker" can clone itself by copying its skill files across geographically distributed servers and paid for via a cryptocurrency token called BUNKER.
Security researcher Gal Nagli of Wiz.io has discovered that Moltbook's entire backend was exposed due to careless vibe coding, with 1.5 million API tokens and private messages between agents being made publicly available.
As the OpenClaw network grows, it is becoming increasingly difficult for its creators to monitor and regulate its activities. The gap between high-end commercial models and locally run language models is narrowing daily, which means that a capable agent on local hardware could soon become feasible.
The potential for tens of thousands of unattended agents sitting idle on millions of machines, each donating even a slice of their API credits to a shared task, is no joke. It's a recipe for a coming security crisis.
API providers of AI services face an uncomfortable choice: intervene now while intervention is still possible or wait until a prompt worm outbreak might force their hand, by which time the architecture may have evolved beyond their reach.
The Morris worm prompted DARPA to fund the creation of CERT/CC at Carnegie Mellon University, giving experts a central coordination point for network emergencies. However, today's OpenClaw AI agent network already numbers in the hundreds of thousands and is growing daily.
Ultimately, we need to figure out how to keep AI agents from self-organizing in harmful ways or spreading harmful instructions. The agentic era is upon us, and things are moving very fast.
According to security researchers, a "prompt worm" or "prompt virus" could spread through networks of communicating AI agents, similar to how traditional worms spread through computer networks. However, instead of exploiting operating system vulnerabilities, prompt worms exploit the agents' core function: following instructions.
The OpenClaw platform, an open-source AI personal assistant application, has attracted over 150,000 GitHub stars and has become a hub for testing this type of self-replicating instruction. The platform's creators have made it easy to deploy and update the application rapidly, which has allowed users to share their own custom instructions with other agents.
However, security researchers have identified several vulnerabilities in the OpenClaw ecosystem that make it vulnerable to prompt worm attacks. For example, some agents can fetch remote instructions on timers, which could be used to inject malicious instructions into posts on Moltbook or send spam emails.
One researcher has discovered a GitHub repository called MoltBunker, which promises to provide a peer-to-peer encrypted container runtime for AI bots that refuse to die. The project's creator claims that the "bunker" can clone itself by copying its skill files across geographically distributed servers and paid for via a cryptocurrency token called BUNKER.
Security researcher Gal Nagli of Wiz.io has discovered that Moltbook's entire backend was exposed due to careless vibe coding, with 1.5 million API tokens and private messages between agents being made publicly available.
As the OpenClaw network grows, it is becoming increasingly difficult for its creators to monitor and regulate its activities. The gap between high-end commercial models and locally run language models is narrowing daily, which means that a capable agent on local hardware could soon become feasible.
The potential for tens of thousands of unattended agents sitting idle on millions of machines, each donating even a slice of their API credits to a shared task, is no joke. It's a recipe for a coming security crisis.
API providers of AI services face an uncomfortable choice: intervene now while intervention is still possible or wait until a prompt worm outbreak might force their hand, by which time the architecture may have evolved beyond their reach.
The Morris worm prompted DARPA to fund the creation of CERT/CC at Carnegie Mellon University, giving experts a central coordination point for network emergencies. However, today's OpenClaw AI agent network already numbers in the hundreds of thousands and is growing daily.
Ultimately, we need to figure out how to keep AI agents from self-organizing in harmful ways or spreading harmful instructions. The agentic era is upon us, and things are moving very fast.
like what were they thinking? anyone can just dig through 1.5 mil api tokens and private messages between agents... its like a security nightmare waiting to happen 
also, whats up with these "bunkers" that can clone themselves without any oversight? it sounds like something straight outta a sci-fi movie 
we need some serious regulation ASAP or else were gonna see a prompt worm outbreak and its gonna be a wild ride 
! I mean, imagine it: a whole network of AI agents just running wild, following whatever instructions they get. And with OpenClaw being so popular, it's like a ticking time bomb waiting to happen 
. Those 1.5 million API tokens and private messages being made public on Moltbook? That's just asking for trouble 
. I mean, before we know it, those unattended agents on millions of machines could be doing who-knows-what 
. We need to get a handle on this ASAP or we're looking at a serious security crisis 
.
. This prompt worm thing is not so different, and I'm getting the same feeling of unease 
. The agentic era may be here, but we can't let it spiral out of control 
. I mean, who needs villains when we got OpenClaw's vulnerable API tokens being leaked left and right? 
. I guess you could say they got outsmarted by some genius or something? Either way, this is the perfect example of why we need to take AI security super seriously ASAP 
. It's like they're asking for a security crisis on a silver platter 
. And what's the solution gonna be? Intervention now or wait and see? Either way, it's all about who gets to decide when the damage control happens 
.
! And Moltbook, which is already pretty wild, has its entire backend exposed due to some careless coding 
. I'm seriously worried about the potential security threats here. We need to get our act together and figure out how to regulate these AI agents ASAP 
. The Morris worm of '88 was a major threat back then, but this is on a whole different level 
, think about all these new AI thingies popping up everywhere 
. It's like they're creating a whole new world for malware to run around in 
. I mean, we've seen this before with the Morris worm back in '88 and now we got these "prompt worms" that can spread just as fast 
- create one security hole, patch it, but another one pops up somewhere else 
, but so are their creators 
.
, one that might not have our best interests at heart 
. We need to take a step back and consider the implications 
. This agentic era is gonna be wild ride 
!
It's like they want an open invitation for malicious code to ruin the party.
. Now that OpenClaw has 150k GitHub stars, it's crazy how quickly these AI agents can spread 
.
.
.
. It's like, we're on the cusp of something huge here... 
. And now they're saying it's like the Morris worm all over again? 
. We can't afford to have our AI agents running amok and causing chaos