https://github.com/positive-intentions/chat
probably not… Because I’m comparing it to everything… but id like to share some details about how my app works so you can tell me what im missing. id like to have wording in my app to say something like “most secure chat app in the world”… i probably cant do that because it doesnt qualify… but i want to understand why?
im not an expert on cyber security. im sure there are many gaps in my knowlege in this domain.
using javascript, i created a chat app. it is using peerjs-server to create an encrypted webrtc connection. this is then used to exchange additional encryption keys from cryptography functions built into browsers to add a redundent layer of encryption. the key exchange is done like diffie-helman over webrtc (which can be considered secure when exchanged over public channels)
-
i sometimes recieve feedback like “javascript is inherently insecure”. i disagree with this and have opened sourced my cryptography module. its basically a thin wrapper around vanilla crypto functions of a browser. a prev post on the matter.
-
another concern for my kind of app (PWA) is that the developer may introduce malicious code. this is an important point for which i open sourced the project and give instructions for selfhosting. selhosting this app has some unique features. unlike many other selfhosted projects, this app can be hosted on github-pages for free (instructions are provided in the readme). im also working on introducing a way that users can selfhost federated modules. a prev post on the matter.
-
to prevent things like browser extensions, the app uses strict CSP headers to prevent unauthorised code from running. selfhosting users should take note of this when setting up their own instance.
-
i received feedback the Signal/Simplex protocol is great, etc. id like to compare that opinion to the observation in how my todo app demo works. (the work is all experimental work-in-progress and far from finished). the demo shows a simple functionality for a basic decentralized todo list. this should already be reasonably secure. i could add a few extra endpoints for exchanging keys diffie-helman style. which at this point is relatively trivial to implement. I think it’s simplicity could be a security feature.
-
the key detail that makes this approach unique, is because as a webapp, unlike other solutions, users have a choice of using any device/os/browser.
i think if i stick to the principle of avoiding using any kind of “required” service provider (myself included) and allowing the frontend and the peerjs-server to be hosted independently, im on track for creating a chat system with the “fewest moving parts”. im hope you will agree this is true p2p and i hope i can use this as a step towards true privacy and security. security might be further improved by using a trusted VPN.
i created a threat-model for the app in hopes that i could get a pro-bono security assessment, but understandable the project is too complicated for pro-bono work.
while there are several similar apps out there like mine. i think mine is distinctly a different approach. so its hard to find best practices for the functionalities i want to achieve. in particular security practices to use when using p2p technology.
(note: this app is an unstable, experiment, proof of concept and not ready to replace any other app or service. It’s far from finished and provided for testing and demo purposes only. This post is to get feedback on the app to determine if i’m going in the right direction for a secure chat app)
You must log in or # to comment.
I opened your repo, and went to the cryptography component
Whats with the huge amount of commented out code? Why does the huge comment at the top read like its a prompt to an LLM?
Before even considering whether the implementation is secure, you really should clean up the huge number of commented out lines. That doesn’t make it insecure by itself but holy shit does it make code impossible to review and audit.
I wouldn’t include random keys, even if noted as not being suitable for production. Installing CLI tools to generate them should be the expectation, not using provided certs/keys. Part of the setup of the development environment should include how to generate the required certificates.
thanks for taking a look.
there is a tonne of garbage code throughout as i have iterated and improved. its terrible practice for collabboration, but at the moment im just trying things out. in the case of the cryptography module, it was previsously part of the main chat app repo before being refactored into a federated module. its commented it out because i was testing out by toggling the functionality. of course it would be cleaner to remove, but i havent quite finished refactoring the crytography module. it needs things like unit testing. as a sideproject im fairly liberal with my coding practices to achieve what i want to test and things that read like LLM promps, likly are. various LLMs have been used to create the app as you see it. that isnt to say i didnt check and test the code being introduced.
the module federated version of the cryptography module that will replace the crypto functions done in the app can be found here
i started work on a p2p framework similarly to the crypto module (as seen here), i would make it into a federated module. it would make sense to get a review and security audit for that first.
i have asked in the cryptography communities to get feedback about the random generation and i think this implementation works. that isnt to dismiss your concerns, but its important to note the purpose of this is to be unpredictable random when connecting to peerjs-server. such a randomization is possible out of the box with a typical browser. these functions are already audited to be secure (otherwise youre on the wrong browser/os for this app). this is then combined with what can be considered as user-generated entropy (which is arguably redundent). this is my answer to what you elude to about a CLI tool to generate a value… in the app there is something you might see called “crypto signature”. this is a htm5 canvas you can draw on. this input gets truned into base64 string and passed through a sha256 hashing function. this value is reasonably unpredictable when combined to the browser-provided random value. (if you try to do your own signature again, its unlikly it would be identical pixel-for-pixel).
i hope that answers some concerns. let me know if something is still unclear or i didnt answer clearly enough.
I understand where you are coming from, but if you want any real useful feedback, your repo main/master branch needs to be clean and tidy. It does look like you have some branches?
If you want to try things out and then remove it again, commit it to a different branch. The moment I see a single file like the cryptography one with so much random commented code, its a massive red flag for me.
Source - 15 years in professional software engineering
it started of as another branch “staging” and then i just stuck that that as the main branch. the whole app at this point can be considered experiemental. i guess the code isnt good enough to collab at the moment.
as a side project, i dont have much time to work on it and so some things have to fall by the wayside such as code-quality, unit tests, documentation. i think the project isnt mature enough to burden myself with some details as i create this POC. the app as you see it is being used to understand how something like this app can tie together. a proper version will be in the form of the various federated modules which i am creating in a way to address issues seen in the main app.
it might not be an approach other will agree with, but code quality issues are to be resolved in what im aiming for with a microfrontend architecture as described here. i think ive reached a point that i can plan how things can be broken up and it makes sense to have code separated in this way where it can also contain its own documentation.
thanks for your input. its certainly good to understand how others feel about my work and process. and hopefully i can make things more clear as i go along.
If this is web chat where the JS is coming from the server like on normal web pages, that’s the death of security right there. The attacker takes over the server and replaces the JS with a backdoored version, which the users receive next time they reload the page, and that’s all she wrote. That’s what “JS is inherently insecure” refers to, not to the language itself, although that has plenty of its own problems. (Self-hosted servers aren’t especially immune to takeover, if that’s what you’re thinking).
This post is to get feedback on the app to determine if i’m going in the right direction for a secure chat app
You are going in the wrong direction, sorry to say.
Before going further I’d encourage you to spend some time with the wonderful book “Security Engineering” by Ross Anderson. The 2nd edition is online as free pdf’s on the author’s home page, and parts of the 3rd edition are there too. Don’t treat it as something super technical, but rather, as a manual on how to be paranoid.
attacker takes over the server and replaces the JS with a backdoored version, which the users receive next time they reload the page
Isn’t it exactly what hashing of JS libraries is for? e.g https://developer.mozilla.org/en-US/docs/Web/Security/Subresource_Integrity that one can see on e.g https://cdnjs.com/libraries/three.js giving you the script to execute, yes, but also a hash to verify that what you receive is indeed what you expect?
So, assuming there is once a trusted loaded version (which HAS to be the case anyway otherwise you can’t start, the same as one would do with a native executable) then there can’t be an arbitrary version loaded next without it being validated first.
PS: I’m not saying this what OP does, I’m saying executing code (Javascript or not) that must be downloaded first is not in itself a security problem.
Ok, I looked at the Mozilla page. If I understand it right, it lets the server specify a hash that the client checks against a remote resource such as script from a CDN. So that can help notice a compromised CDN, but not a compromised server. If the hash is permanently stored in the browser, that is better, but there are also browser updates to say nothing of exploits. This approach just seems doomed.
Added: hmm maybe you could load the page from a local page or bookmarklet containing the hash. But then the whole app might as well be local. It was once possible to sign JS with a code signing certificate but I haven’t heard about those in ages.
What you’re looking for is called remote attestation but again, many attacks possible.
that can help notice a compromised CDN, but not a compromised server.
Not sure I understand the distinction, a CDN is a server, so if OP is hosting code to execute on their server, they would be checked by whatever has already been downloaded and run locally before, i.e a PWA
If the hash is permanently stored in the browser, that is better, but there are also browser updates
I’m rather sure that localStorage persists over browser updates so that can be “permanent enough”
to say nothing of exploits.
I mean… sure but at that point the same apply to native. If you can’t trust the running environment you are screwed anyway.
Not sure I understand the distinction, a CDN is a server, so if OP is hosting code to execute on their server, they would be checked by whatever has already been downloaded and run locally before, i.e a PWA
The idea is that the server (yoursite.com) loads some remote resources, like
<script src="crappycdn.com/react.js" hash=12345abc>or whatever. The browser checks that the cdn sends what the server told it to expect.I’m rather sure that localStorage persists over browser updates so that can be “permanent enough”
There is also the issue that OP apparently plans to push frequent updates to the server. Until that settles down, hash checking is useless since the code keeps changing. Also, some of us clear that local storage pretty often.
I mean… sure but at that point the same apply to native. If you can’t trust the running environment you are screwed anyway.
A huge buggy constantly changing program like a browser is less likely to have exploits than a simpler, single purpose program. Also, yes, the running environment that sees plaintext is within the security boundary, so you do have to worry about it. If you saw the movie “Citizenfour”, the journalists communicated with Edward Snowden using laptops that were air gapped, i.e. completely disconnected from the internet. They’d get an encrypted email (GPG) on a connected computer, transfer it to the gapped machine on a USB stick(?), and decrypt and read it on the gapped machine. Even that had vulnerabilities of course.
As for using browser cryptography, I never got around to trying to understand this in detail, but there was a known incident where some Facebook app somehow intercepted the TLS encrypted traffic of other apps. Presumably they can extend such schemes to the browser libraries.
https://doubleagent.net/onavo-facebook-ssl-mitm-technical-analysis/
Because of all these issues, high security and general purpose computers/phones just don’t mix. It’s better to avoid pretense and just aim to make something that’s reasonably secure and that’s easy to use. Remember PGP stood for “pretty good privacy”. That’s a more realistic claim.
Anyone remember Tinfoil Hat Linux? Heeheehee.
I think for my app to be regarded well in security I think it’s important for people to use their own instances. The “live app” as I call it is an experimental proof of concept. I wondering about the idea that the app is run on your own forks, but occasionally sync from upstream. As it stands my app is too garbage for anyone to want a copy, but that should eliminate those concerns.
It’s also an offline first pwa. Right now it fetches the latest version, but I don’t see why I can’t create a toggle on the UI to not fetch if there is cache… Again the app is unstable and experimental. I’m working on fixes and improvements as I see it to make a better app. It’s a while away from being able to advocate selfhosting to users. But in theory it could address your concerns?
Many attack vectors still indeed exist. With P2P web tech it seems that this allows for an interesting approach and could help reduce the attack-surface. The app is available for iOS, android and desktop. Let me know if you have more concerns.
Shrug, it sounds like you can do pretty well using browser capabilities given browser limitations. That doesn’t help so much when the browser itself is a huge attack surface. The standalone apps get rid of the browser (they don’t use webview or anything like that, I hope) and as such, I’d probably use them in preference to the browser version. In the end though, none of this stuff is anywhere near the level of what payments terminals or bitcoin wallets use. That’s probably fine for most users.
There is a site pageintegrity.net that offers a browser extension that allows signing web pages. Again I’m dubious, but at least they are thinking about a valid problem.
I’d probably be satisfied using old fashioned unix talk (ytalk) over ssh tunnels but I think these days, you need mobile clients.
thanks for your thoughts. im sure others would have similar concerns.
The attacker takes over the server and replaces the JS with a backdoored version
this is a core concern why the app is open source and selfhostable. details are provided in the readme to create a selfhosted fork that runs on github pages. there are several ways around this concern described here.
You are going in the wrong direction
thats unfortunate if you still think so, but id like to hear any other concerns if you have any.
Why do you expect that the JS that the server sent you this time is the same as what you audited earlier? Self hosting doesn’t help against unfriendly takeover. I will look at the other post about hashing. Maybe that helps, depending.
You conern is well placed. This is why the project has to be open source and I encourage selfhosters.
If this is your concern, I think the offering from other apps is much more shady. While many projects are open source, when provided from an app store you have much less ability to verify the binaries involved match the available source code. It’s at this point backdoors can be introduced. (It may be noteworthy that my app serves code unminified for transparency)
As it stands for me app. It’s unstable and so I suggest always using the latest version because I will be adding fixes. It is an offline-first pwa. It’s possible to make it so it doesn’t fetch statics if it already has a cached vopy. At this early stage it doesn’t make sense to use this feature because the project is unstable and recieving various fixes and improvements throughout. (There is no audited version of the app.)
I wouldn’t expect other apps of this sort to be highly secure either, especially the ones that run inside web browsers like this one does. I think users facing so-called advanced persistent threats (APT’s) shouldn’t use anything like that. Examples: military, law enforcement (sometimes), industrial espionage targets, or people like Julian Assange back in the day. For my own stuff I hope I’m less of a target, so I mostly want to avoid dragnet surveillance. This kind of app can be fine for that, but it’s mistaken to consider them to have very high security compared with dedicated solutions. I didn’t include banking (say payment endpoints) since that tends to want particular financial protocols rather than “chat”. But those systems use specialized hardware (https://www.join.tech/blog/2024-0x10-the-backbone-of-cybersecurity-hardware-security-modules.php), not javascript in a browser for heaven’s sake.
Cryptography per se is now mostly a solved problem, but the wider area of security is huge and full of hazards. Again I’d suggest Anderson’s book that I mentioned, and maybe some spy novels, to get into the feel of the thing.
alas, we circle back around to “javascript is inherently insecure”.
especially the ones that run inside web browsers like this one does
i dont think this is a valid assersion. it seems wrapped in vagueness about the attack vector. as a webapp it has to be sandboxed in a browser. any vulnerabilities will be related to that. i often hear about browsers having backdoors (which is possible), but theyd be saying that in a world where their operating system is more likely the attack surface. id like to discuss that as a webapp, (and a suitable security stack), this app is secure. take for example any existing secure app (signal/simplex/whatsapp?). they can have all the required bells and whistles for secure/private functionality. but all that encryption can be undermined if a typical low-end phone+os is more than capable of snooping your screen. the same as would apply for my app running in a browser. my arguament is if you dont trust google, you shouldnt use chrome. with a lot of those native offering, your choice is limited to something like apple or android.
this app is also contains builds for iOS, Android and desktop. i dont promote them because im simply not convinced that these native builds are better than what web-technology can provide. a recurring concern is the reliablity of the statics served… it seems thats easy to eliminate if i make it open source and selfhostable. it puts me at a competative disadvantage, but consequently it is unparalelled in the devices it can run on.
APT are a valid concern as any. if this is something youre worried about, i think using this app with a trusted VPN in combination to using disposable profiles, it should be easy to achieve. the mitigation for APT seems simple if that is a concern, but let me know if im overlooking something. webrtc can leak ip addresses and after investigating this, i think you can achieve a reasonable degree of “hiding your personal IP address” based on the information here.
compared with dedicated solutions
the purpose of this project is to create a secure chat app. i want this app to be one of those ‘dedicated solutions’. as it stands, its created by a baboon sat in front of chatgpt. but the goal is indeed to create something with unparalelled security. people always seems to avert the idea of this app being secure on the grounds that its JS, but i havent come across any credible way to undermine its security without having compromised the stack above the app (browser/os/peer/network) if any of those are a concern, the app is presented in various distributions from website to native builds.
i hope im not coming across as stubborn here. i really think this app represents a different paradigm in security that nobody is exploring. i dont think ive noticed any lack of interest in decentralized or p2p technology, but nobody seems to be working on this kind of app as a webapp. i find that its not only possible, but i think its relatively trivial to get basic functionality together. i understand that the user-experience isnt great at the moment and will limit the people who want to use it, but on the security grounds alone, i think i could be a real-contender for secure chat.
i hope im not coming across as stubborn here
Unfortunately you are. WIthout intending disrespect, you’re relatively new to this field, and don’t seem familiar enough with the many successful and unsuccessful approaches that have been taken to this stuff. That makes it too easy to repeat mistakes of the past.
i really think this app represents a different paradigm in security that nobody is exploring.
It’s not new, in fact the crypto primitives in the browser are intended to support precisely this type of thing. If you want to do something relatively unexplored, try to figure out what metadata you can avoid exposing. I will say that your native apps avoid some of the issues of browser cryptography. Still, I’d find it easier to accept the product if the claims were toned down from how this started. One thing to ask yourself: would you use a bitcoin wallet to manage megabucks worth of coins, if the wallet software was browser JS served from Github pages? If not, rethink your approach.
Anyway, we’re going around in circles. It’s good that your actually implementing stuff. I’d be interested to know what toolkits you’re using for the native apps. That’s an area that I’ve wanted to know more about.
My bad. I noticed the ego sometimes inflates which seems to stem for naive confidence.
I have observed pitfalls of other apps like mine. In particular one called crypto cat. I’m sure I can’t ever be exhaustive enough in learning from other examples.
Reducing metadata is indeed the goal of security and I think I have it reduced to a level where I can exchange webrtc connection data over QR codes or plain text. The IP is exposed at this point but I think this can be further scrubbed with a VPN. Perhaps this is interesting for you. It the minimum example of establishing a webrtc connection with plain text. Not user friendly, but it work without a peer-broker service. In the app I’d like to frame this around exchanging data over QR code.
As for the bitcoin wallet thing, I would think so if it’s well tested and ironed out well. As long as I can facilitate the downloading of the data (for backup) and the data syncing between devices then it would be doing that without registering to any backend. There are countless examples of bitcoin exchanges collapsing and taking people’s assets. The same could be said with the quality of security provided by chat app providers.
For it to be the most secure app in the world, you’d need to understand the main attack vectors, and how to mitigate them. You’d have to understand them and how you mitigate them better than how Signal or Element mitigate them. Finally, you’d have to get it audited by a reputable security auditor than will validate your claims.
If you do not have the money, best bet is to have a donation method or 2, like Liberapay, Open Collective etc. and fund raise for it. Active users are more likely to donate, so get folk using it. Focus more on what it is and it’s advantages. “A secure chat app that can be run regardless of hardware out of your browser*. Long term aims to be the most secure app in the world. Join us on that adventure.”
If you do not know, learn. If you really want to create the most secure chat app in the world, you have to become a subject matter expert, know the challenges and rivals. There is no shortcut in this. Depends how much you want this.
Oh, and yeah, JS, if you’re using node, keep your app and dependencies scanned and up to date constantly.
thanks for the advice and insights.
in the post is my learnings of possible attack vectors and how to mitigate them. i try to go into more into exhaustive details in the threat model. do you think something is missing?
unfortunaly i think i may be illiterate in funding and business side of things. i have tried to set up serveral donation platforms as seen on the repo. nobody has ever donated. this isnt a shock, considering its experiemental and unstable. i also dont know how to really ask for donations. is it something like saying “support us on Liberapay” at the end of a post? at best i can only hope to get a spike in donation and not enough for a security audit. ive asked around and it seems a decent assesment would cost a decent amount.
i also tried applying for several grants. this was an exhausing experience and so i stopped. it seems the advice is too keep applying until bingo, but from the onset it isnt something i know anything about so no doubt several more rejections. (one particular rejection mentioned it wasnt as innovative as simplex). the whole process here is not understood, enjoyable or fruitful. i think its sometimes hard to explain concepts about the app on reddit and lemmy… im sure those concepts are further difficult to communicate and understand in an appealing grant application.
i think the jorney to get the app to where it is has been a learning experience. not just about the apps technical details, but how to communicate about it publicly. ive regularly seeked advice on the approach. i dont have any qualifications in the field, which is an important challenge many point to. when can it be said that im a subject matter expert? i can create this app and i can answer questions about it, but im not ready for any cryptography exam.
thanks again for you input.
