Microsoft mishandling example.com | Hacker News Reader

Microsoft mishandling example.com

tinyapps.org

249 pointsmrled89 comments

>Microsoft's Autodiscover service misconfiguration can be confirmed via curl -v -u "email@example.com:password" "https://prod.autodetect.outlook.cloud.microsoft/autodetect/d...":

Hold up, does this mean outlook sends your full credentials to Microsoft when you try to set up an outlook account? I'm sure they pinky promise they keep your credentials secure, but this feels like it breaks all sorts of security/privacy expectations.

It's more common than you might think. I know of at least one popular email client that stores your credentials on their servers to enable features like multi-account sync and scheduled sending.

I would expect such a feature to use end-to-end encryption for the data, so that only the user can see the credentials. It does, right? Right?

>>multi-account sync and scheduled sending

>I would expect such a feature to use end-to-end encryption for the data

How would "end-to-end encryption" when such features by definition require the server to have access to the credentials to perform the required operations? If by "end to end" you actually mean it's encrypted all the way to the server, that's just "encryption in transit".

> If by "end to end" you actually mean it's encrypted all the way to the server, that's just "encryption in transit".

This is what Zoom claimed was e2ee for a little while before getting in trouble for it.

This is what Google also claims as end to end encrypted in their Gmail end to end thing. Many people including me mentioned this in the comments.

https://news.ycombinator.com/item?id=45458482

Its entirely their end to their end encrypted. You don't get any privacy.

Use our new open source (modification and redistribution not permitted) app to exchange end-to-end encrypted (from your client to our server) messages with your friends! Having all your data on our service protects your data sovereignty (we do not provide for export or interop) by guaranteeing that you always have access to your full history! Usage also protects your privacy (we analyze your data for marketing purposes) by preventing unscrupulous third parties from analyzing your data for marketing purposes.

If we had competent regulators this sort of blatant willful negligence would constitute false advertising.

Do you mean Spark? I get why they need to do it that way but I also hate that they have to do it that way because it sucks for privacy.

Yeah, Spark. Shame because I really liked their client, but I refused to use it anymore after I realized what they were doing.

I bought a hardware password manager a while back and the bulk load tool sent all your creds to a cloud service. I have not used it since, and sent the manufacturer a nasty note.

It was the Ethernom Beamu, company now defunct.

I think outlook is pretty much a saas product these days.

What gave it away, the intrusive ads in your free inbox the last ten years,

or the “See Plans and Pricing” on the homepage?

Christ, my poor grandmother…

Most likely, and nobody cares.

Already many years ago I remember installing a firewall on my phone and noticing in surprise that Outlook was not connecting at all to my private mail server, but instead only sending my credentials to their cloud and downloading messages from there.

The only Android mail client not making random calls to cloud servers was (back then) K-9 Mail.

Yeah since the Windows 11 2023h2 update.

Not just that, the new outlook app makes Microsoft a complete man-in-the-middle for your email account.

https://www.xda-developers.com/privacy-implications-new-micr...

And? Do you think Gmail is end to end encrypted?

I am talking about the fact that the new default email client on Windows will hand over all your email credentials to Microsoft. This has nothing to do with Gmail.

Oh you mean even if you don't use Microsoft's email? Now I get it.

My bank isn't end to end encrypted either, but that doesn't mean it's suddenly ok for Microsoft (or any other company) to suddenly start MITMing my online banking connections.

Adding a bunch of middlemen that also see the data increases the risk.

I think the concern is that it copies the emails of your non-Microsoft accounts that you added to the Outlook app, over to Microsoft servers

This seems like it would completely break any attempt to track access from unauthorized users or devices — any IT department using a backend other than Microsoft’s would need to pretend that all access from MS’s servers is safe.

In response to discovering this any competent IT department would immediately move to ban the use of any offending apps and blacklist the MS servers from the relevant backends. Also I guess rather than drop the connections ideally you would want to accept the initial request, record the provided credentials, and then lock said account because the credentials have clearly been compromised and the user is now known to be making use of a banned app.

It’s also the case that, of the major cloud providers, one of them is quite notably poor at securing its own systems. If I were a company that cared about security, I would not want Microsoft holding credentials to my system.

They store passwords and proxy everything at the same time they’re pushing OAuth, authenticators, passkeys, etc. for their own services. Everyone should have revolted when they bought Acompli and started doing this kind of thing.

I am so glad people are finally noticing and complaining about this. It's the same reason I won't use Spark or Superhuman. Those are neat services, but I can't abide storing the creds to perhaps the most security-sensitive service I use to a cloud provider. If they get hacked, then the attacker can access my email account, send phishing emails to my contacts, read and respond to password reset requests they make to other online services, etc. It would be disastrous.

No, I'll keep my credentials stored and used locally, thanks.

So like Cloudflare for email.

Basically everything microsoft makes that touches http will send your username and your password to any server that asks for Basic Authentication.

It looks like Microsoft Edge had the _ability to disable_ this added in 2020 or 2021, but it isn't currently the default and the Group Policy unintuitively only applies to unencrypted HTTP Connections.

>Basically everything microsoft makes that touches http will send your username and your password to any server that asks for Basic Authentication.

Are you talking about NTLM hashes? It's a weak hash, but not the same as "sending your password". The biggest difference is that even a weak hash can't be reversed if the password has high enough entropy.

Not necessarily, the server can say it only supports basic auth and….

I don't think there's any evidence that windows sends cleartext passwords. The whole reason why NTLM is a thing is to avoid sending cleartext passwords.

Outlook appears to be

The 'https://' disagrees with your 'sending clear text passwords' statement.

It’s clear text to the receiving server, which is what we’re talking about, not one way hashed.

yes, I meant to type hash. Not that it matters as even 10yr old integrated GPUs are enough to brute force 8 or 9 character NTLM(or any variant) passwords in a few hours. Not that you need to with Pass The Hash.

> Hold up, does this mean outlook sends your full credentials to Microsoft when you try to set up an outlook account?

Not just an “outlook account” - any account in outlook, with default settings at least.

I run a mail server, mainly for me but a couple of friends have accounts on there too, and a while ago one friend reported apparently being locked out and it turned out that it was due to them switching Outlook versions and it was connecting via a completely different address to those that my whitelists expected sometimes at times when they weren't even actively using Outlook. Not only were active connections due to their interactive activity being proxied, but the IMAP credentials were stored so the MS server could login to check things whenever it wanted (I assume the intended value-add there is being able to send new mail notifications on phones/desktops even when not actively using mail?).

> but this feels like it breaks all sorts of security/privacy expectations.

It most certainly does. The behaviour can be tamed somewhat, but (unless there have been recent changes) is fully enabled by default in newer Outlook variants.

The above-mentioned friend migrated his mail to some other service in a huf as I refused to open my whitelist to “any old host run by MS” and he didn't want to dig in to how to return behaviour back to the previous “local connections only, not sending credentials off elsewhere where they might be stored”.

I might be misremembering but i think it even copies all of your mails to their servers.

Always has been.

I think the curl -u switch just requires the password field to be filled, there obviously isn't a legit user account test@example.com with a password of password either at microsoft or at the Japanese imap server.

>I think the curl -u switch just requires the password field to be filled

Yeah you're right, if you don't specify the password (eg. -u user), it prompts you for it

>there obviously isn't a legit user account test@example.com with a password of password either at microsoft or at the Japanese imap server.

But presumably the fact it's there at all suggests it's a required parameter? Maybe "password" is just a placeholder, but it's unclear based on the command line transcript alone.

See also: Windows 11 telemetry

> Microsoft's Autodiscover service misconfiguration can be confirmed via curl -v -u "email@example.com:password" "https://prod.autodetect.outlook.cloud.microsoft/autodetect/d..."

Wait, does their autodetect send email and password to their servers, instead of just domain???

See replies to a similar question here (in case you haven't already): https://news.ycombinator.com/item?id=46732623

Autodiscover has always been an interesting security problem. I wrote this years ago:

https://lolware.net/blog/2020-09-02-autodiscover-circus/

Not surprised. They used to have training material incentivizing professionals to use .local as TLD for Active Directory realms. Thats a reserved domain for Multicast DNS.

Working on Linux automation systems we would need to make sure to disable anything related to Avahi in our images otherwise name resolution would fail for some customers.

My company used .local for EVERYTHING. I took it as normal at the time, until I got into problems with VMWARE products.

Support patiently explained .local is reserved for something else and kindly provided Wikipedia links.

They never responded why they used .local in their docs, trainings, webinars they provided, though :)

My impression is that Ballmer IE6 era Microsoft didn't gave a shit about standards.

Is standard you are talking about is Multicast DNS https://www.rfc-editor.org/rfc/rfc6762 from year 2013?

I’ve worked with hundreds of customers that use .local internal domains and vmware, what issues are you describing?

Things from docs making it into production is insidious. There were some early Sun docs that referenced a 129.9.0.0/16 network. Some helpful contractor in my locality, specializing in local government work, configured several police, fire, and city governments with that subnet internally back in the 90s. A few of them are still running that way today. I remember running into some oddball behavior with the Teredo adapter in Windows 7 that I traced back to it behaving differently because the PC's IP address didn't fall into RFC1918 space.

Makes me remember the 192.1 addresses that were all over at one place I worked. "Um you know that is a valid internet address right?" "Yeah, but the guy who originally set the systems up was confused about the private address space, used the wrong one and we don't want to break anything so are not going to change it"

Good times.

Haven't they been telling people to do that since before it became reserved? If so, the problem is more that you can't "reserve" something that's already in wide use, and mdns should've used something like .mdns.

It's like when .dev became a gTLD, knowingly breaking a bunch of setups for a mix of vanity and a cash grab. Obviously dropped the ball on the engineering side.

Seems more a reason to never use stuff you don't actually control and are reserved for future purposes. Everyone knew who was in charge of DNS TLDs and that while they were being at first conservative in how many they assigned, they reserved the right to assign as many as they wanted.

But also, yes Microsoft documentation used .local before mDNS reserved it, and IIRC Microsoft was also involved in suggesting it for mDNS as mDNS came out of the multi-company standardization efforts from Apple's Bonjour. That said, my impression of most of that documentation from that time is that it was incorrectly using .local as a fake TLD where they should have been using .example or .example.com and also pointing people to the RFCs that those were not valid choices in a real setup. A problem with such documentation is that it is too easy to take literally. A follow up problem was sort of the "accidental security through obscurity" benefits of using non-globally resolvable addresses becomes "best practice" through essentially stubbornness and status quo (related to all the recent rediscussions on HN about NAT44 is not a firewall except by accident and you can have very good firewalls that aren't NAT44).

> my impression of most of that documentation from that time is that it was incorrectly using .local as a fake TLD

When setting up Active Directory on Windows Server 2003, there was a note in the wizard that explicitly called out .local as a domain suffix that would prevent DNS lookups from hitting the public internet, which many people (myself included) took as an endorsement.

I feel like I remember this in 2008 as well. I could certainly be misremembering.

> Haven't they been telling people to do that since before it became reserved

If you actually try to find an evidence for this (even time traveling to 2015 before the great wipe of most pre-Vista docs) you wouldn't find a confirmation for this. What you would find is what the official docs always recommended the root domain to be an official bought one on the public internet. And this excludes .local.

Usage of .local for AD predated mDNS. That advice stopped with the advent of mDNS in favor of 'corp.<registered_domain>.<tld>'.

The original Windows 2000 guidance for AD was corp.example.com, from my recollection. The silly .local thing (which does predate mDNS) happened as a result of the Small Business Server refresh for Active Directory.

Just a guess but why do I get the feeling it’s because someone who setup sei.co.jp in Azure Entra (aka Azure AD) some how managed to add/claim the domain “example.com” against their companies tenant.

It’s clearly not using the DNS records for discovery because they don’t exist, the only other option I can see is some weird fall through or hard coded value and it seems like an odd one to pick.

Where does sei.co.jp comes from? Why Microsoft would use that domain in the first place?

It's not really the domain but the registration in the MS Office Cloud. If you query who owns example.com mail you get that company.

I'm willing to bet they were the first user to try and add example.com to their Outlook account, and MS then just assigned it to them without verifying they own the domain.

> The domain has a null MX record (indicating it doesn't accept email)

Not quite true, SMTP will use the A record if there is no MX.

In this case, "null MX record" means MX exists, but does not specify a valid server:

   $ host -t mx example.com
   example.com mail is handled by 0 .

Senders should not fall back on the A record in this case.

That’s why example.com states “Avoid use in operations”, not only that could create unnecessary traffic for them as well as leak information as in situations like this.

Yeah, it feels more like a safety net than something you should purposefully use

Why do you need to send a password when using their Autodiscover API? Would Outlook send the respective passwords for each email account to Microsoft?

I suspect they try to login and reverse engineer the IMAP config.

curl -u just requires the field to be there, I suspect. No authentication takes place. You can send any password and the output doesn't change.

This is why I never use these IANA-reserved domains like .test, .example, .invalid, .localhost.

I always make up some impossible domains like domain.tmptest

Otherwise you're one DNS "misconfiguration" away from sending dev logs and auth tokens to some random server.

> Since at least February 2020, Microsoft's Autodiscover service has incorrectly routed the IANA-reserved example.com to Sumitomo Electric Industries' mail servers at sei.co.jp, potentially sending test credentials there.

.example is probably far safer than example.com.

https://www.akamai.com/blog/security/autodiscovering-the-gre...

According to it, it seems that if someone registers autodiscover.com then example.com lacking autodiscover.example.com will make Outlook try checking if autodiscover.com has an entry.

It's just a braindead system.

It's all fun and games until Donuts buys .tmptest for some reason.

brb, just filing paperwork to apply for the .tmptest gTLD /s

I suspect you'd download a car.

$100K

$227k just to apply, and another few hundred thousand in legal, compliance, and contracting to reach delegation.

Source: I'm on the board of dotMeow and wrote the financial plan

Would that really make a difference in this case? It's a configuration error / bug in Microsoft's discovery server, they could have a fallback that goes "any unknown address, return this .jp address".

It so happens that in this very specific case your obviously bad choice didn't make anything worse, that doesn't make it a good choice.

"Aha, the defective trucks only cause injuries to people who have their hands on the wheel at highway speeds, but I've never bothered holding the wheel at high speed, I just YOLO so I wouldn't be affected"

If people had used IANA's reserved TLDs they too would be unaffected because although Windows will stupidly try to talk to for example autodiscover.example that can't exist by policy and so the attempt will always fail.

And then you fire off 100k emails, they all bounce, and your mail service shuts you off...

As others have pointed out, using 'tmptest' works until someone buys tmptest -- unlikely, but people will buy anything these days.

I always use the ISO-3166 "user-assigned" 2-letter codes (AA, QM-QZ, XA-XZ, ZZ), with the theory being that ISO-3166 Maintenance Agency getting international consensus to move those codes back to regular country codes will take longer than the heat death of the universe, so using them for internal domains is probably safe.

The correct one to use is .internal

It is reserved by ICANN since 2024-07-29.

https://en.wikipedia.org/wiki/.internal https://www.ietf.org/archive/id/draft-davies-internal-tld-00...

You can also safely use .home .corp and .mail as those have been explicitly rejected by ICANN on the basis that they would cause widespread naming conflicts. I have my devices configured to redirect queries against .home to a local nameserver, leaving .local open for avahi.

I gather this has little to do with “example.com” and more to do with any domain that doesn’t have an autodiscover subdomain.

Nice to see tinyapps.org is still alive.

Now that example.com is hidden behind Cloudflare, how does the public know who is controlling the origin servers

The IPv4 for example.com used to be 93.184.216.34

Was there an announcement somewhere

The old nameservers for example.com still have not updated their zone files

   dq a example.com 199.43.135.53

This is the same company that mishandled the Office brand (abandoned it) and is mishandling the Xbox brand (what even is an Xbox anymore?). Are we surprised?

NSA probably. Gives them plausible deniability.

Maybe some of their targets did use example.com for some probing, and the NSA had a hand in Sumitomo Electric Industries' mail server.

Reading the article, there is a huge flaw in the autodiscover protocol by Microsoft.

https://www.akamai.com/blog/security/autodiscovering-the-gre...

According to it, it seems that if someone registers autodiscover.com then example.com lacking autodiscover.example.com will make Outlook try checking if autodiscover.com has an entry.

It's just a braindead system.

Reply on news.ycombinator.com