had to edit title to have it fit length limit. paper title is "More is Less: How Group Chats Weaken the Security of Instant Messengers Signal, WhatsApp, and Threema"
absolutely. raising the noise floor would make it harder for the adversary to pick up valuable signals. but why bother with complicated jamming setups that require you to cover a very wide spectrum, at high power, while taking care of not interfering with your own equipments, and still have signals going out that with sufficient means and motivation, a sophisticated attacker may still intercept (we're pretty good at extracting valuable signals from noise) when you can just put on your RFP to stick the lot in a faraday cage, use optical connectivity inside it, forbid wireless HIDs, buy shielded monitors and shielded phones, and put a couple of scary bald men in front of the door.
air gapping removes connectivity to the wider world, and a computer system (as in, a system of computers) can totally have value without being connected to anything that is outside of the organization.
no need to resort to hyperbolic alien computers that are anyway easily hacked by plucky heroes with a taste for ascii skulls.
you're correct in that of course we may want to insert and retrieve data from a secure zone. but when those types of systems are necessary, "manual human intervention" is far too blurry. manual data entry/retrieval by vetted people ? a HUMINT problem the organizations that tend to require those systems are familiar with, and have adequate procedures to deal with. same with printouts and assorted "low tech" issues. inserting a USB key and ensuring that the whole chain doesn't compromise your very valuable system is a far more complex problem that you probably do not want to deal with if you can avoid it. there is a reason the "usb gadget give away" and "lost usb key in the parking lot" are a staple of flashy pentests demos. it works remarkably well, and is remarkably hard to defend against without enacting procedures that make the life very much harder on everyone. (try suggesting epoxying the USB ports of every single computer and servers in a medium to large company to its CISO)
like another commenter remarked. it is truly the operational challenges of ensuring your system says properly isolated from the rest of the world that are hard to deal with. cutting a cable is definitely not the end of it.
i'm sure people that deal with high assurance environments and are more qualified than i am to answer specifics have their share of stories of creatively compromised isolated environments.
you'd be surprised.
when specifications get passed around in the endless maze of defense affiliated sub contractors and sub sub contractors, they may be implemented in all kind of creative ways.
and i entirely agree that the operational issues of enforcing the air gap are by far the most challenging cost in this kind of environment.
the culture of business environments is by and large ill suited to the kind of strict enforcement these environment require.
if RF leaks are part of your threat model, then i would definitely not class them as "proximity attacks". without adequate shielding, antenna black magic and a rented apartment on the other side of the street work perfectly well to capture signals of interest.
that is not what an air gap is. if at any time there is an exchange of data between the air gapped hosts/network and a non air gapped one, there is no air gap anymore. and yes, that includes an analyst taking a sample to an internet-connected host.
side channel exfiltration techniques like "fansmitter"[1] and "diskfiltration"[2] or the well known TEMPEST attacks are more interesting challenges in air gapped designs than relying on a very basic security boundary violation. (and solved in very much the same way, by having the secure terminals be as dumb as bricks as possible, with extremely limited known inputs and outputs, preventing the insertion of unclassified electronics in the secure zone through policy, and enforcing it through physical searches, RF hardening the zone with a faraday cage, absence of windows, etc... the "Technical Specifications for Construction and Management of Sensitive Compartmented Information Facilities"[3] is a great ressource for those interested in such designs)
incidentally, there are awfully little true air gapped networks (which are not ICSes and/or assorted single purpose hosts/networks whose absence of network connectivity is not a primary design feature) in the wild. they're operationally heavy precisely because of their nature, and outside of very specific situations, their security is not perceived as cost efficient by management, for good reasons.
i am, i get people to use it, and it works well. but superuser2 is right. to be honest, i think we're touching to fundamental limitations in the git model that gitlab cannot solve without becoming rational team concert, and catering to a fundamentally different public.
if locked code review is coming in, i'll be all over it. just let me select multiple approvals, and nested approvals.
it's an immensely useful feature. once again, the more i can define a process in software, the better it is for most non-western, non-modern-management companies.
yeah i read the scope. but since the border between scheduling, configuration management, monitoring etc... is extremely blurry, with a high number of interconnection between systems, and lacking any kind of standard contract between them, i'm more wary about the introduction of these features.
firefox has a plugin that works fine from what i've heard from the others. i still rely by default on totp for generic dual factor auth, since i can provide already working solutions from web to ssh authentication. however, yes for web applications, some clients went the u2f options, and the users overwhelmingly prefer it. since u2f is getting very quickly traction in products (case in point, gitlab) and the user experience is so simpler for a similar level of security, it'll probably become my go to in the near future.
gitlab is gearing up to be the one-stop-shop from development to production. i really like that. unifying the toolset developers and ops have to understand is really a boon for everybody involved.
merge restriction as a feature may sound silly, but in my experience, when driving organizations toward CI/CD, especially with "mature" organizations, having the computer say no instead of a human is a major facilitator in adoption of good practices. now we just need to protect tests in tree. (currently, i tend to favor tests as a git submodule, so we can prevent chronic offenders from altering/commenting tests that should pass)
plus U2F integration, so now i won't hear anymore how "terribly slow" it is to have to whip out a phone to grab a TOTP code...
pyrotherapy is a real thing, and has been used very successfully in the treatment of the later stages of syphillis. of course, it only works because syphillis is really temperature sensitive, which isn't the case of the pathogens concerned.
we're in the process of getting all the zone files we can to reduce the amount of DNS requests we have to do, but the real kicker is whois databases, for example AFNIC asks for 10K€ to get access to a copy of the database...
a lot has changed since our last Show HN, and i guess it was time to share these change with you.
for those who weren't here the first time, we try to catch domain squatting using a bunch of techniques we already used when doing it manually, but in a purely automated fashion. your root domain (think 'facebook', 'twitter', 'ycombinator', etc...) gets through blenders that generate variants, that we'll gather info on.
we now have all the basics in place so i can confidently call it production ready. free accounts are still the same deal, one domain, five TLDs, all present and future generators, whois and dns resolver, plus a few more still in the oven, and one run per week, which should be enough if your needs are not massive and/or specific.
you also now have paying options for people with more intensive needs, either timing wise (down to one run per day, and one run per 4 hours will be a possibility once i'm confident 1: we can handle it, 2: it can actually provide value in the real world) or number-of-tlds-wise.
notifications! yeap i know it's basic, but we now send you a mail when a run is complete, so you don't have to bother reloading waiting for that progress bar to reach 100%. a few client asked us about sms notification, but i'm not sure about multiple notification channels yet.
so what's next :
we have a bunch of stuff that stayed on the backburner while i was working on making the production as autonomous as possible (complete CI/CD stack, built with chef, openstack heat, jenkins, the whole shebang) and the other dev was working on ironing the kinks
existing when interacting with horrible protocols like whois (for the sake of everyone's sanity, i really hope rdap (http://about.rdap.org/) gets traction) or misbehaving dns, or just plain old bugs. we're now bringing them back on the front of the workbench.
parking detection.
this one is simple, and everybody will get it, but i noticed a large number of parked domains in resolution runs, so they'll be marked as such.
automated phishing detection.
this has been a major demand so i'm prototyping a CV system (ab)using ghost.py and opencv to see if i can get something that has a reasonable false positive rate.
malware detection.
a smaller demand because it's already well covered by other products. for the moment paying accounts get access to google safebrowsing, and i have a bunch of threat exchange APIs access ready to enter the quorum. there's a lot of datasharing between those, so i don't want to generate false positives because of data sharing. i have also been working slowly on PR for cuckoo sandbox that'll help me launch fleets of sandboxes in various configuration (hopefully i'll have enough variants that the client is able to more or less choose the one that correspond to its production environment to try and catch targeted attacks)
keyword prediction based on the root domain.
we have a keyword generator that can generate domain variants, think 'cheap-brand' for 'brand', but if you're like me you probably can't think of a lot of those (i had good success asking marketing guys for ideas). once again AI to the rescue, i'm tracking which keywords had the most 'success' in finding resolving variants, which means once i'm able to establish lexical domains i'll be able to offer everyone a 'most likely keyword for this domain' help to feed the generator.
an API!
at the very beginning, when dinosaurs roamed the earth and the iphone 5s the cool product of the year, squatmon was just a very large and very ugly python script i used in various recon engagements. as we decided to slap a shiny web interface on it and share it with others, we didn't take the time to make an API the first class citizen and the web interface just the reference client implementation.
this is a mistake we intend to correct, so any person with an account, free or otherwise can integrate any of the functionality they have access as a part of something bigger. (i have written an example postfix milter that's too terribly slow to be used in production, but can participate in the spam score of an email for example)
if you have any question, want to report bugs, or anything really, don't hesitate to contact me, my email address is in my profile.
edit: i'm terrible and i said one run per month on the free account. it's of course one run per week
yes apparently i broke OCSP. thanks for the tip i'm fixing it now.
edit: apparently the error lies around certificate transparency and us lacking a public CT log. the security of the connection itself is not impacted, but i'm looking into it.
my edit shouldn't change the meaning.