I absolutely despise Firebase Firestore (the database technology that was “hacked”). It’s like a clarion call for amateur developers, especially low rate/skill contractors who clearly picked it not as part of a considered tech stack, but merely as the simplest and most lax hammer out there. Clearly even DynamoDB with an API gateway is too scary for some professionals. It almost always interfaces directly with clients/the internet without sufficient security rules preventing access to private information (or entire database deletion), and no real forethought as to ongoing maintenance and technical debt.
A Firestore database facing the client directly on any serious project is a code smell in my opinion.
Securing the db is more of an ops thing.
Peak Vibe Coding results.
while True:
Jesus Christ
You know that’s not the Tea code, but the downloader, right?
They’re also not using requests very efficiently, so who knows.
Other reports state the Tea backend was Vibe Coded: https://www.ainvest.com/news/tea-app-data-breach-exposes-72-000-users-ai-generated-code-security-lapse-2507/
Sure, it might be, I’m not saying it isn’t. All I’m saying is: the screenshot shows the code someone wrote to download the images. It’s not part of the Tea codebase.
who’d have thought that javascript and client side programming was incredibly susceptible to security flaws and deeply unsafe
As much as I dislike JavaScript, it isn’t responsible for this. The person (or AI) and their stupidity is.
but it didn’t help; it was basically the gasoline
When i tried making a website with gemini cli it did deadass use string interpolation for sql queries so everything is possible
Robert’); DROP TABLE Students; –
aw bobby
who’d have thought that being shitty programmer was incredibly susceptible to security flaws and deeply unsafe instead of javascript
No, it must be JavaScript that is the problem
principal_skinner.jpg.exe
Microsoft defender identified a malware in this executable.
Wow. It actually identified something?
This reminds me of how I showed a friend and her company how to get databases from BLS and it’s basically all just text files with urls. “What API did you call? How did you scrape the data?”
Nah man, it’s just… there. As government data should be. They called it a hack.
ah yes, the forbidden curl hack
Believe it or not a lot of hacking is more like this than you think.
Security by obscurity.
If I was a hacker, I would just get a job as a night cleaning person at corporate office buildings. And then just help myself to the fucking post-it notes with usernames and passwords on them.
I think that’s less about “hacking” and more about modern day devs being overworked by their hot-shit team lead and clueless PMs and creating “temporary” solutions that become permanent in the long run.
This bucket was probably something they set up early in the dev cycle so they could iterate components without needing to implement an auth system first and then got rushed into releasing before it could be fixed. That’s almost always how this stuff happens; whether it’s a core element or a rushed DR test.
modern day devs being overworked
And then there is meningspunktet.dk which had all the time in the world to do whatever they wanted, and even get their hosting paid for by a university. They still leaked everyones email, phone, full legal name and location on day one and only fixed it because I pointed it out.
Thanks for the writeup - I can’t believe they claimed anything about keeping data safe while building the website so poorly…
Social engineering is probably 95% of modern attack vectors. And that’s not even unexpected, some highly regarded computer scientists and security researchers concluded this more than a decade ago.
When the technical side reaches a certain level of security, the humans become the weakest link.
*if
We reached that part a long time ago.
Clearly the authors of this app did not. Hence “if.”
This has been the case for 40+ years. Humans are almost always the weakest link.
we built this shit. thus we are always to blame.
The percentage is closer to 75% than 95%.
I work in security and I kinda doubt this. There are plenty of issues just like what is outlined here that would be much easier to exploit than social engineering. Social engineering costs a lot more than
GET /secrets.json.There is good reason to be concerned about both, but 95% sounds way off and makes it sound like companies should allocate significantly more time to defend against social engineering, when they should first try to ensure social engineering is the easiest way to exploit their system. I can tell you from about a decade of experience that it typically isn’t.
https://www.infosecinstitute.com/resources/security-awareness/human-error-responsible-data-breaches/
You’re right. It’s 74%.
https://www.cybersecuritydive.com/news/clorox-380-million-suit-cognizant-cyberattack/753837/
It’s way easier to convince someone that you are just a lost user who needs access than it is to try to probe an organization’s IT security from the outside.
This is only going to get worse with the ability to replicate other’s voices and images. People already consistently fall for text message and email social engineering. Now someone just needs to build a model off a CSO doing interviews for a few hours and then call their phone explaining there has been a breach. Sure, 80% of good tech professionals won’t fall for it, but the other 20% that just got hired out of their league and are fearing for their jobs will immediately do what they are told, especially if the breach is elaborate enough to convince them it’s an internal security thing.
Yes social engineering can be incredibly effective. I completely agree, but there is a bit of an obsession with it these days and imo it’s over indexed, because at the end of the day the type of social engineering detailed in that report typically just provides access.
In some cases, the target is important enough and has enough organizational power that accessing the network as them is sufficient, but that’s not often the case. What that means is that in those other cases social engineering (which in that report you cited is often just phishing) is providing, typically, internal network access. An attacker will have to move through the network and exploit software typically to continue their attack. There are many points in this chain that the weakness lies in software or configuration. If effort was placed on making those systems better it would likely see better results than hyper focusing on the social engineering, which is significantly more difficult to stop, especially with all of the things you mentioned on the horizon.
My point is then that even if it is a part of 74% of breaches, according to Verizon, it’s not necessarily sufficient and is often paired with software level exploits.
And I know this because my company does plenty of red teaming, and we use social engineering but at the end of the day the more interesting result comes from a software exploit or just abusing a weak configuration.
Shodan lists 100’000s of publicly accessible security cameras.
Many years ago, I discovered that my then-employer’s “home built” e-commerce system had all user and admin passwords displayed in plaintext at home/admin/passwords.
When I brought this to the attention of leadership, they called the “developer” in and he said “oh, well, that’s IP locked, so no one on the web can access it!” When I pulled it up on my phone, he insisted my phone was on the work WiFi, despite it being clearly verifiable that was not the case. (The same work WiFi that had an open public connection, which is the one my phone would have been on, if it were on it…)
He did fix that, but many other issues remained. Eventually a new COO hired someone competent as his ‘backup’, replaced our website and finally suggested he pursue other employment opportunities before he could no longer voluntarily pursue them. (There was concern he might sabotage.)
Disabling index and making the names UUID would make the directory inviolable even if the address was publicly available.
Security through obscurity never works.
It’s not security through obscurity in this case. The filenames can’t be obtained or guessed through brute force. At least not with current technology or processing power…
Security through obscurity is when you hide implementation details.
Saying that my suggestion is security through obscurity is the same as telling that ASLR is security through obscurity…
Sounds like a good case for brute forcing the filenames. Just do the proper thing and don’t leave your cloud storage publicly accessible.
While proper security is better, you’re not gonna brute force UUIDs.
As long as you’re not rate limited, you absolutely could.
A UUID v4 has 122 bits of randomness. Do you know how long that would take to brute-force, especially with network limitations?
It taking a long time doesn’t make it an impossibility. The fact that it has a limit of 122 bits, in and of itself, makes the possibility of a bruteforce a mathematical guarantee.
By this logic, all crypto is bruteforcable, on a long enough timeline.
A 122 bit random number is 5316911983139663491615228241121378303 possible values. Even if it were possible to check 1 trillion records per second, it would take 168598173000000000 years to check all the UUIDs and get the info on all the users. Even if every human on earth signed up for the app (~8 billion people), and you wanted to just find any one valid UUID, the odds of a generating a UUID and that being valid in their DB is basically 0. You can do the math your self following the Birthday Paradox to determine how many times you would need to guess UUIDs before the probability that any one UUID is valid against a population of the whole world is greater than 50%.
You should read into the NSA’s Translator. Granted, it’s relatively outdated with shifting text algorithms, but for a very long time (about half a century), it was able to bruteforce any key, regardless of length, in under an hour.
For all practical purposes, it’s impossible.
It’s not, though. And thinking that it is impossible is why DES, for example, was “translatable” by the NSA for decades. Never assume something is impossible just because it’s difficult.
You cannot!
I cannot. But the bruteforce is a mathematical guarantee.
And has nothing to do with my proposition.
Can’t be done.
AI just enables the shit programmers to create a greater volume of shit
I’ll tape this to my office door.
What was the BASE_URL here? I’m guessing that’s like a profile page or something?
So then you still first have to get a URL to each profile? Or is this like a feed URL?
It’s a public firebase bucket
That should be criminally negligent.
Oh Jesus
🤦♂️
Possibly from the decompiled APK. 404media reported that they found the same URL as the posted one in the APK (archive link).
I remember when a senior developer where i worked was tired of connecting to the servers to check its configuration, so they added a public facing rest endpoint that just dumped the entire active config, including credentials and secrets
That was a smaller slip-up than exposing a database like that (he just forgot that the config contained secrets) but still funny that it happened
That’s not a “senior developer.” That’s a developer that has just been around for too long.
Secrets shouldn’t be in configurations, and developers shouldn’t be mucking around in production, nor with production data.
I would have put IP address access restrictions on that at the very least. I may have even done something like that more than once for various tools in the past.
That way it acts completely open to people (or other servers) in the right places and denies all knowledge to anything else.
Does anyone have a source for this?
The original article is paywalled (I mean, registration-walled?), this summary is not
404 Media reported that 4chan users claimed to be sharing personal data and selfies from Tea after discovering an exposed database.
Even the best models fine tuned for coding still have training that was based on both good and bad examples of programming from humans. And since it’s not AGI but using probability to generate the code, you’re going to get crap programming logic dependent on how often such things were used and suggested by humans to other humans. Googling for an answer on how to code something pulls up all sorts of answers from many sources, but reading through them, many are terrible. An LLM doesn’t know that, it just knows that humans liked some answers better than others, so GIGO.
Gorilla In Gorilla Out?
Sounds like a good time
Giraffe In Giraffe Out
Gorilla In Giraffe Out
That would be the real trick.
Fantastic for building BaaS apps
Bullshit as a Service?
Bananas as a Service :)
bananas in pyjamas
I wonder if their data is poisoned by below average Dev. I mean if your test subjects are met or below Dev and mad Ethel lost 20% efficiency imagine what you can do to good dev
Not below average dev necessarily, but when posting code examples on the internet people often try to get a point across. Like how do I solve X? Here is code that solves X perfectly, the rest of the code is total crap, ignore that and focus on the X part. Because it’s just an example, it doesn’t really matter. But when it’s used to train an LLM it’s all just code. It doesn’t know which parts are important and which aren’t.
And this becomes worse when small little bits of code are included in things like tutorials. That means it’s copy pasted all over the place, on forums, social media, stackoverflow etc. So it’s weighted way more heavily. And the part where the tutorial said: “Warning, this code is really bad and insecure, it’s just an example to show this one thing” gets lost in the shuffle.
Same thing when an often used pattern when using a framework gets replaced by new code where the framework does a little bit more so the same pattern isn’t needed anymore. The LLM will just continue with the old pattern, even though there’s often a good reason it got replaced (for example security issues). And if the new and old version aren’t compatible with each other, you are in for a world of hurt trying to use an LLM.
And now with AI slop flooding all of these places where they used to get their data, it just becomes worse and worse.
These are just some of the issues why using an LLM for coding is probably a really bad idea.
Yeah, once you get the LLM’s response you still have to go to the documentation to check whether it’s telling the truth and the APIs it recommends are current. You’re no better off than if you did an internet search and tried to figure out who’s giving good advice, or just fumbled your own way through the docs in the first place.
You’re no better off than if you did an internet search and tried to figure out who’s giving good advice, or just fumbled your own way through the docs in the first place.
These have their own problems ime. Often the documentation (if it exists) won’t tell you how to do something, or it’s really buried, or inaccurate. Sometimes the person posting StackOverflow answers didn’t actually try running their code, and it doesn’t run without errors. There are a lot of situations where a LLM will somehow give you better answers than these options. It’s inconsistent, and the reverse is true also, but the most efficient way to do it is to use all of these options situationally and as backups to each other.
Yes, it can be useful in leading you to look in the right place for more information, or orienting you with the basics when you’re working with a technology that’s new to you. But I think it wastes my time as often as not.
That’s implying that the quality of information from other sources is always better, but I’m saying that’s sometimes not true; when you’re trying to figure out the syntax for something, documentation and search engines have failed you, and the traditional next step would be to start contacting people or trying to find the answer in unfamiliar source code, sometimes a LLM can somehow just tell you the answer at that point and save the trouble. Of course you have to test that answer because more often than not it will just make up a fake one but that just takes a few seconds.
There are some situations I’m going back to search engines as a first option though, like error messages, LLMs seem to like to get tunnel vision on the literal topic of the error, while search results will show you an unintuitive solution to the same problem if it’s a very common one.
whether it’s telling the truth
“whether the output is correct or a mishmash”
“Truth” implies understanding that these don’t have, and because of the underlying method the models use to generate plausible-looking responses based on training data, there is no “truth” or “lying” because they don’t actually “know” any of it.
I know this comes off probably as super pedantic, and it definitely is at least a little pedantic, but the anthropomorphism shown towards these things is half the reason they’re trusted.
That and how much ChatGPT flatters people.
Yeah, it has no notion of being truthful. But we do, so I was bringing in a human perspective there. We know what it says may be true or false, and it’s natural for us to call the former “telling the truth”, but as you say we need to be careful not to impute to the LLM any intention to tell the truth, any awareness of telling the truth, or any intention or awareness at all. All it’s doing is math that spits out words according to patterns in the training material.
Didn’t expect this much. I don’t think about tuto example being weighted heavier. This make sense.
Not a big fan of the wording here. Plenty of skilled programmers make dumb mistakes. There should always be systems in place to ensure these dumb mistakes don’t make it to production. Especially when related to sensitive information. Where was the threat model and the system in place to enforce it? The idea that these problems are caused by “shit programmers” misses the real issue: there was either no system or an insufficient system to test features and define security requirements.
I can tell you exactly what happened. “Hey Claude, I need to configure and setup a DB with Firebase to store images from our application.” and then promptly hit shift+tab and then went to go browse Reddit.
nothing was tested. nothing was verified. They let the AI do its thing they checked in on it after an hour or so. once it was done it was add all, commit -m “done”, push origin master. AI doesn’t implement security stuff. there was zero security here.
I have found the exact same type of bug shown here probably over a dozen times, most of those long before AI was writing code.
I found a bad programmer!
I found someone who hasn’t yet made their big dumb mistake. Give it time.
I’ve dodged the bullet for 20 years, now. I guess i had better get cracking
You’ve probably already made your Big Dumb Mistake, it just hasn’t been triggered yet.
Or, you just weren’t there any more when it triggered.
