I’ve heard anecdotes of people using an entirely internal domain like “plex.example.com” even if it’s never exposed to the public internet, google might flag it as impersonating plex. Google will sometimes block it based only on name, if they think the name is impersonating another service.
Its unclear exactly what conditions cause a site to get blocked by safe browsing. My nextcloud.something.tld domain has never been flagged, but I’ve seen support threads of other people having issues and the domain name is the best guess.
I'm almost positive GMail scanning messages is one cause. My domain got put on the list for a URL that would have been unknowable to anyone but GMail and my sister who I invited to a shared Immich album. It was a URL like this that got emailed directly to 1 person:
Then suddenly the domain is banned even though there was never a way to discover that URL besides GMail scanning messages. In my case, the server is public so my siblings can access it, but there's nothing stopping Google from banning domains for internal sites that show up in emails they wrongly classify as phishing.
Think of how Google and Microsoft destroyed self hosted email with their spam filters. Now imagine that happening to all self hosted services via abuse of the safe browsing block lists.
if it was just the domain, remember that there is a Cert Transparency log for all TLS certs issued nowadays by valid CAs, which is probably what Google is also using to discover new active domains
It doesn’t seem like email scanning is necessary to explain this. It appears that simply having a “bad” subdomain can trigger this. Obviously this heuristic isn’t working well, but you can see the naive logic of it: anything with the subdomain “apple” might be trying to impersonate Apple, so let’s flag it. This has happened to me on internal domains on my home network that I've exposed to no one. This also has been reported at the jellyfin project: https://github.com/jellyfin/jellyfin-web/issues/4076
Well, that's potentially horrifying. I would love for someone to attempt this in as controlled of a manner as possible. I would assume it's possible for anyone using Google DNS servers to also trigger some type of metadata inspection resulting in this type of situation as well.
Also - when you say banned, you're speaking of the "red screen of death" right? Not a broader ban from the domain using Google Workplace services, yeah?
This reminds me of another post where a scammer sent a gmail message containing https://site.google.com/xxx link to trick users into click, but gmail didn't detect the risk.
I'm kind of curious, do you have your own domain for immich or is this part of a malware-flagged subdomain issue? It's kind of wild to me that Google would flag all instances of a particular piece of self-hosted software as malicious.
- A self-hosted project has a demo instance with a default login page (demo.immich.app, demo.jellyfin.org, demo1.nextcloud.com) that is classified as "primary" by google's algorithms
- Any self-hosted instance with the same login page (branding, title, logo, meta html) becomes a candidate for deceptive/phishing by their algorithm. And immich.cloud has a lot of preview envs falling in that category.
BUT in Immich case its _demo_ login page has its own big banner, so it is already quite different from others.
Maybe there's no "original" at all. The algorithm/AI just got lost among thousands of identically looking login pages and now considers every other instance as deceptive...
I’ve heard anecdotes of people using an entirely internal domain like “plex.example.com” even if it’s never exposed to the public internet, google might flag it as impersonating plex. Google will sometimes block it based only on name, if they think the name is impersonating another service.
Its unclear exactly what conditions cause a site to get blocked by safe browsing. My nextcloud.something.tld domain has never been flagged, but I’ve seen support threads of other people having issues and the domain name is the best guess.
I'm almost positive GMail scanning messages is one cause. My domain got put on the list for a URL that would have been unknowable to anyone but GMail and my sister who I invited to a shared Immich album. It was a URL like this that got emailed directly to 1 person:
https://photos.example.com/albums/xxxxxxxx-xxxx-xxxx-xxxx-xx...
Then suddenly the domain is banned even though there was never a way to discover that URL besides GMail scanning messages. In my case, the server is public so my siblings can access it, but there's nothing stopping Google from banning domains for internal sites that show up in emails they wrongly classify as phishing.
Think of how Google and Microsoft destroyed self hosted email with their spam filters. Now imagine that happening to all self hosted services via abuse of the safe browsing block lists.
if it was just the domain, remember that there is a Cert Transparency log for all TLS certs issued nowadays by valid CAs, which is probably what Google is also using to discover new active domains
It doesn’t seem like email scanning is necessary to explain this. It appears that simply having a “bad” subdomain can trigger this. Obviously this heuristic isn’t working well, but you can see the naive logic of it: anything with the subdomain “apple” might be trying to impersonate Apple, so let’s flag it. This has happened to me on internal domains on my home network that I've exposed to no one. This also has been reported at the jellyfin project: https://github.com/jellyfin/jellyfin-web/issues/4076
3 replies →
Well, that's potentially horrifying. I would love for someone to attempt this in as controlled of a manner as possible. I would assume it's possible for anyone using Google DNS servers to also trigger some type of metadata inspection resulting in this type of situation as well.
Also - when you say banned, you're speaking of the "red screen of death" right? Not a broader ban from the domain using Google Workplace services, yeah?
1 reply →
This reminds me of another post where a scammer sent a gmail message containing https://site.google.com/xxx link to trick users into click, but gmail didn't detect the risk.
Chrome sends visited urls to Google (ymmv depending on settings and consents you have given)
Yes, my family Immich instance is blocked from indexing both via headers and robots.txt, yet it's still flagged by Google as dangerous.
I'm kind of curious, do you have your own domain for immich or is this part of a malware-flagged subdomain issue? It's kind of wild to me that Google would flag all instances of a particular piece of self-hosted software as malicious.
G would flag _some_ instances.
Possible scenario:
- A self-hosted project has a demo instance with a default login page (demo.immich.app, demo.jellyfin.org, demo1.nextcloud.com) that is classified as "primary" by google's algorithms
- Any self-hosted instance with the same login page (branding, title, logo, meta html) becomes a candidate for deceptive/phishing by their algorithm. And immich.cloud has a lot of preview envs falling in that category.
BUT in Immich case its _demo_ login page has its own big banner, so it is already quite different from others. Maybe there's no "original" at all. The algorithm/AI just got lost among thousands of identically looking login pages and now considers every other instance as deceptive...
I have my own domain, and Immich is hosted on an "immich" subdomain.
1 reply →