Though it would be good if there was an open source communications platform that would allow the public to engage with politicians in a formal and constructive way. Alas, so much disparity in solutions that it often irks me.
https://ec.europa.eu/digital-single-market/en/projects Be great if the interface was better for that site, though google didn't find anything of note.
I can think of many comparable situations in other countries (some EU ones as well) in which the person finding the issue would of very easily been locked up.
The actual bug was thanks to a long-standing bug in python's standard email.utils library, which finally got fixed: https://bugs.python.org/issue34155, combined with insufficiently-defensive coding and testing on my side. (I wrote the auth code in question).
It does not really say that the app is "super not secure". Just that people make mistakes, and it's not even shameful the way they reacted to it.
My favorites was the "special email inbox for lawyers".
So, I am somewhat hesitant to let the EU develop a chat client.
I hated working there because I felt like a parasite just making the world worse. The in-hosue IT of our customer would probably have done a better job if they had just been allowed to hire some guys instead of having to go through an easily gameable procurement procedure. Not because they would have been more competent, but because they would have not have had perverse incentives.
I mean, we've got this nice secure login system for government sites: DigiD, which works fine. So you need that to submit your taxes. A few years ago, it turned out you needed any valid DigiD to submit your taxes, not necessarily your own. There are tons of other sites that cost many millions, and resulted in an atrocious user experience. And some projects simply failed after going way over budget.
Estonia has a similar solution, but "Smart-ID" was created after PKI was fully rolled out, it's terrible compared to that. Centralized, unverifiable, stores secrets insecurely, does some cryptographic bullshit. The two solutions somehow seem very related.
But my point is that even if DigiD is secure, that's still not going to help you when the tax service uses it incorrectly.
I haven't heard any such criticism about that system before, so it would be interesting to hear. I've been using it for years, it's easy to use and there haven't been any security problems with it. The company that developed it is trying to export it to other countries as well.
It doesn't utilize a hardware-backed secure secret storage on Android (~80% of marketshare in EE if I remember correctly). This means vital key material is unsafe from any compromise, just a plain file somewhere. This means that they have to go into a lot of effort to detect clones, which is also certainly not infallible. Cloning or compromising an ID-card of Mobile-ID is way way harder with it's audited security.
The technical overview describes how key is generated on the device and half of it is sent to AS SK's servers. There's no security benefit in that (they can't really verify you've deleted their half of the key from your device), except that now AS SK has half of your private key. In general their technical overview gives the feeling that like they want to pull wool over people's eyes, it's IMHO a bad sign.
It also totally lacks any privacy, every login you do is logged and counted. SK AS shouldn't mandatorily have that metadata.
It relies completely on their centralized servers, they go down, your authentication and signing goes totally down. Neither service owners or their clients can do anything about that. Identity mustn't be like that.
It's totally proprietary, thus their claims about security can't be easily verified, and it's limited to their blessed OSs (and versions). If they go bankrupt or similar, the identities of all those people are in jeopardy.
It's also expensive for server administrators and thus has a high barrier of entry. Identity and security shouldn't be behind a paywall for neither side. The web has LetsEncrypt and now a lot more sites are protected, the web is better off, SK AS has done the opposite.
In comparison, ID-card's PKI is usable for free, supported in every system (that can do TLS basically), entirely FOSS if you wish, the hardware is much more secure (EAL6+), it doesn't mandatorily leak metadata, doesn't rely on a centralized proprietary server to work.
Then use that for the competition to get the contract.
Preselection is hot garbage, as proven time and time again. Of course mechanism to mitigate the upfront cost of participating in such a competition should be implemented but in the end it should turn out cheaper/more worthwhile since a working MVP already exists by time of selection.
And anyway most costs are incurred during the lifetime of the project, after you proved yours is the best solution, with endless change requests, delays that require additional money, etc. The companies bidding usually have far more experience at syphoning money than the client has at holding on to it.
The development is also more expensive than in other areas since there is need for documentation.
And then there is a flaw in the process: Producing the software has to be done via some bidding process. For writing the requirements documents they already rely on companies, which can add clauses making it hard for competitors, after that the lowest bidder wins. In case there is a competitor and not some big player like T-Systems as the only one the offers typically are "optimistic" meaning that at some point in time the budget will run out and extra budget has to be granted (or the project would fail, which doesn't look good and they'd have to restart the whole process, thus wait another two years) and there a "smart" company can find quite a few extra costs. ("Oh wait, you updated from windows XP to 10, we have to redo a lot and then re-twst and re-certify")
It's a game some companies can play quite well.
That all said: Software is expensive. Gathering requirements is hard. Bureaucracy, which among other things tries to prevent corruption, causes extra work.
When, say, the national Police asks companies to come up with sofware to manage contact details for all employees, you could come up with "Install and Configure SomeOpenSourceCRM on a VPS" for €3000.
Whereas typically some Enterprise Consultancy quotes a factor hundred of that to "integrate" it in a crappy way.
I've worked in such projects, for governments. The overhead is ridiculous. the NIH-syndrome is rampant. The specs and "but this is how we do things"-requirements are cut in stone.
Very often a evening of tuning some Open Source tool could suffice; But either some rule disallowing the language/server/deploy-speed, or some manager disliking it because last year they decided that from now on everything must be Java+XML, or you need at least three weeks of meetings before you are allowed to even start.
This is not a joke. I've had a simple off-the-shelve Rails tool cut down, because "we don't allow dynamic languages". I've had projects delayed with 3+ weeks because "we require a deploy-window to be announced 21 days in advance" and I've worked as only developer on a project with four(!) managers managing me.
That is why government projects are expensive: a catch22: "Because they are".
> we don't allow dynamic languages
I would translate as "we have now experience in running and maintaining those platforms, which causes issues"
> require a deploy-window to be announced 21 days in advance"
This sounds a bit like ITIL or similar processes in oalce, which are there to ensure the systems are in defined state.
> with four(!) managers managing me.
Usually those should have different roles. Like technical oversight, communicating with the actual users and getting their requirements, then the IT management for operations and than oversight (variance exists, but lots of things, different stakeholders and as they spent public money they have to do extra documentation of processes for accountability reasons)
While I also feel the temptation to yell "I could do it for 1/100th of that" at a screen when I read/hear about these projects, just basic probability suggests that I'm probably wrong. It's me (with as little enterprise" experience as I could manage) vs a whole lot of people that don't strike me as particularly stupid and who have lots of incentives to keep costs down.
When you try to buy software like this through conventional state procurement mechanism you'll inevitably waste a lot of money, but it would be really cheap to "grow" it through supporting an environment where that can happen. Why doesn't Werner Koch have a position in academia? Third party funding of whole departments is commonplace and is always a blind bet trusting that the grantees will come up with something rewarding.
[0] (In German) https://www.heise.de/newsticker/meldung/Open-Source-Bundeswe...
Forking a piece of software whose market moat is 100% network effects is not easy.
As a citizen within EU I would rather see them donate money to the Signal Technology Foundation instead. OTOH that could be controversial, so perhaps it would be better if citizens and companies would donate directly instead.
[0] Signal conducts the hiring and interviews, EU pays the salary
The EU's structure is also a bit different than that of nation states, with cabinet members and parliament rather limited in their ability to make spending decisions. Anything serious needs approval by the council, i. e. the heads of state.
And just now is probably a bad time for new expenditures, as Brexit will severely cut into revenue.
Signal does the job, doesn't store data, not even metadata, and can be used immediately.
But the code is right there for you to read it and re-use it, you just won't get far building your own network with one user and demanding everybody else switch over.
Isn't that just a promise? Also, even then, it is based on your contacts, which are seen by google.
You can use a different contacts app, you don't have to give all your information to Google. My contacts are managed by a Nextcloud instance.
As far as I know OWS does not mention this anywhere on their site nor on their program -- aren't the issues with usability of other programs and lack of sane defaults (such as with gpg) often given as an argument by signal supporters on why you should prefer it?
That being said, is that even possible? I admit that I am not too familiar with how Android phones work. Signal requires the google play services in order to work, right? Is this not enough for google to see your information?
Signal also released a WebRTC version that doesn't depend on Google Play Services if that floats your boat.
Your phone number is sent to Signal's servers during sign-up and it uses the conventional SMS service to "close the loop" and prove this number is under your control. Having signed up you can use a PIN to lock the number to you so that anyone without that PIN can't do the "new phone" dance (this expires if you stop answering PIN questions correctly)
If you choose to do so a digest of your contact's phone numbers can be sent to Signal for them to match against the set of (also digested) numbers of Signal users so they can tell you who has Signal enabled.
Whether you choose to give your contacts to Google, to Facebook, to Apple or whoever is up to you and outside Signal's control.
Signal does let you create an encrypted profile, and then your device can tell other people's devices the keys to look at the profile if you want to allow that. You don't have to use a profile or trust anybody else if you don't want to. Signal doesn't learn the keys (unless I guess you deliberately sent them those keys) so they can't read the profile.
Unlike many of its competitors Signal's messages can't be read by Signal, in most cases this includes who sent them (Signal's "Sealed Sender" means in most cases if you correspond with someone the indication of who sent them a message will be encrypted such that they can tell you sent it but Signal only knows it was someone they authorised to send them messages). When you attach images Signal avoids learning how large the images are exactly, and if you use a service like GIPHY to add typical meme images like Stephen Colbert eating popcorn Signal double-proxies this so that they don't learn which GIF you used, and GIPHY doesn't learn who used it.
Edit: Fixed name of GIPHY. Huh.
I can look it out for myself but there won't be any point as they can simply run different code on their servers.
> If you believe that despite precautions the source code won't match what actually runs on your phone
On their servers
Also what precautions? As far as I know their binaries are not reproducible.
> this expires if you stop answering PIN questions correctly
After a week if I remember correctly.
> a digest of your contact's phone numbers
> also digested
A hash? This does not protect against anything. There are much less than 2^32 active mobile phone numbers per country. It would be trivial to brute-force it.
> Whether you choose to give your contacts to Google, to Facebook, to Apple or whoever is up to you and outside Signal's control.
The point is that someone* other than you will be able to see the metadata. It does not matter if it is Signal or not.
OK.
That's true for literally all services. Do you expect to be able to walk into the server rooms and dump the binaries to inspect them?
> Also what precautions? As far as I know their binaries are not reproducible.
The client builds are. Reproducible server builds don't tell you anything about what is running.
The point is that your client should not send any information which you expect to keep private to their services. It is the exact reason that we use e2ee rather than just tls for chats.
> The client builds are
Not fully, see https://signal.org/blog/reproducible-android/
> Reproducible builds for Java are simple, but the Signal Android codebase includes some native shared libraries that we employ for voice calls (WebRTC, etc). At the time this native code was added, there was no Gradle NDK support yet, so the shared libraries aren’t compiled with the project build.
> Getting the Gradle NDK support set up and making its output reproducible will likely be more difficult.
Yes. And Signal achieves this better than all the other major options, given the number of footguns in the other tools.
If you are concerned about the client builds then run a decompiler. It's not hard. People have been auditing binaries for ages.
https://www.aclu.org/open-whisper-systems-subpoena-documents
You can also disable contact backup on Android.
Developing a custom app would mean throwing tax money into a black hole, generating an inferior product that nobody would use (among other things due to the complete lack of network effect). Supporting an existing, good open-source product (and possibly spending grant money on _that_) makes a lot more sense, and is a practical solution.