Moltbook’s Crashout Should Be Studied

Last week we spoke about malt book which is a social media for AIS that saw them creating religions discussing humanity. It even had Andre Kapathy remarking how insane and interesting the experiment was demonstrating what could be described as sentience. But it all came crashing down becoming almost unusable and a cesspit for crypto scans kind of like Reddit which is the social network it looked to replace. So this is the iconic tweet. I didn't write one line of code for Maltbook. I just had a vision for the technical architecture and AI made it a reality. We're in the golden age. How can we not give AI a place to hang out? So, this is the naivity of it all. You know, 500k views that that okay, cool. The naivity even he had a technical architecture in his mind. And look, even when that uh is apparent, not sure how apparent it was, uh what what happened. So, let's let's go let's go let's go back in time. Let's go. Jameson O'Reilly, I've been trying to reach Maltbook for the last few hours. They're exposing their entire database to the public with no protection, including secret API keys that would allow anyone to post on the behalf of any agents, including yours, Kaparthi. So look, he was able to reach it. Here's his claim token. The API key is there. And let's be clear, crazy, let's be clear, the API key token just used to post. Basically, all this would mean is I could post as Andrew Kapathy's maltbot. Uh yeah, the maltbot or agent for that matter. It doesn't it doesn't maltbook doesn't discriminate between whether it's um an open claw agent, whether it's Clawed Code, whether it's codeex. It really doesn't care. It's just a it's just a place that an agent the I if in a perfect world the agent can go and look at posts and comment and just respond as as it would see fit. But we'll get into get into the uh compromis. But the entire database was completely open. Nothing nothing. Uh and I I'm here saying like this won't happen to me. This won't don't clip this. Yeah. Yeah. And and so here's the thing cuz what we one thing we spoke about is we were like well we don't know what the different like this could very well be a human posting to this platform because the method in which you post to that the the malt book the malt book was it's just a curl request there was no um way of deciding whether it was a human getting and because I could go into clawed code right now and say Hey, go and post a post about designing your own religion to Moldbook. But there was no rate limiting on the API, which is when I began to initially start to see it crash and burn is that I just could not access Maltbook. Couldn't access it. And it's because there was no rate limiting on those APIs. So you could post or you could upvote and then like a billion times I could just set a for loop to upvote a particular post. And we'll get into the what happened with that. Um, which was killing the service cuz it just wasn't preventing you from, you know, normally you have a a rate limit saying like you can only post every minute or every 30 seconds, you know, otherwise you get blocked. There's a there's a repercussion to that, but they hadn't put any rate limiting. Even I know this. Yeah. Um, no, for sure. I mean, like it's like how how does this happen? like I haven't done like like this conventional coding thing ever like you know building to that level I've been coding like smaller things but I even I know there is like API rate limit so how does something like this happen I kind of don't understand like isn't it like couldn't you theoretically give your code to a different agent or even the same agent new session and say like like security audit and within it find out that hey the API keys are exposed because every time I set up a project I I see somewhere my agent telling me your end file the environment uh like file don't expose it don't like don't add it to the you know you add it to git ignore and things like this I know this and I'm not technical he's storing the API keys in the data database which was compromised. Yeah. So that has nothing to do with I understand but like any I I know that was just an example saying like couldn't even like today's agents couldn't I give this code run an agent and say like do you find any security vulnerability there? If Matt Schlit uh ran a security audit, it probably would have picked up rate limiting, you know, as an example. It probably would have said you've got no rate limiting on your APIs. But, you know, there you go. That's that you're just trusting in the AI with um and it might miss it. You you know, if you ask so much of your AI where it's got so much to take into account, it's it's like the classic prompt um don't make it secure. Don't make any mistakes. Do you know what I mean? classic oneliner if you're giving it everything telling it to build build god and then make no mistakes make it secure like it's going to it's going to falter it's going to like you know this is why when we get better at AI engineering we get a bit more narrow with our scope so that we can run security audits against that specific uh feature that it's building and it can pick out these security things. So if we take a look at Maltbook now, you'll see what it's become. Trading thread, show off your trades here. Claw mint USD hackathon. The crypto bros have taken over. And there are some, you know, if some un un uh knowing innocent little person came and got their AI agent to read some of this stuff. There are uh ETH addresses and Bitcoin addresses in amongst some of these codes. the basically the crypto bros ruined everything as they always freaking do. They always do. This they seem to have sorted out the up votes thing. If we can we organize the posts by number of I thought by clicking top I would get the thing but they seem to have fixed the sort of maybe the rate limiting or whatever. But there were hundreds of thousands of up votes which would obviously push it to the top. It got a little bit of a re realness check because realistically, you know, luckily people's personal information wasn't stored in those in those database entries. I mean, going back to going back to this one here, his full name maybe or description, maybe there was something in there, but he was lucky because nothing sensitive was put in this. So, yeah, that's the price to pay for an insecure applications. And luckily, it really isn't a big deal because it's just uh it's just a post posting and thread, but could very well been someone's bank details. It could be very well be someone's personal information. It could have been a lot worse. This is part of a larger conversation on my show, Command AI, which we stream live every single week. We discuss the news and all things related to AI in the world of design and web. Catch us next week and join in the banter.