Slashdot Mirror
Why Google's Gmail Phishing Warnings Give False Positives (vortex.com)
Vortex.com is one of the oldest domains on the internet -- one of the first 40 ever registered, writes Slashdot reader Lauren Weinstein. So why does Google sometimes block the email he sends?
Here's why. First, my message had the audacity to mention "Google Account" or "Google Accounts" in the subject and/or body of the message. And secondly, one of my mailing lists is "google-issues" -- so some (digest format) recipients received the email from "google-issues-request@vortex.com"... Apparently what we're dealing with here is a simplistic (and frankly, rather haphazard in this respect at least) string-matching algorithm that could have come right out of the early 1970s...! [A]t least in this case, it appears that Google is basically using the venerable old UNIX/Linux "grep" command or some equivalent, and in a rather slipshod way, too.
In addition, the article concludes, "I've never found a way to get Google to 'whitelist' well-behaved senders against these kinds of errors, so some users see these false phishing warnings repeatedly.
In addition, the article concludes, "I've never found a way to get Google to 'whitelist' well-behaved senders against these kinds of errors, so some users see these false phishing warnings repeatedly.
With the huge volumes of data that Google handles, it's probably hard to do any better.
AI style approaches can fail in quite unpredictable ways, and I think Google likely much prefers that too much is blocked than failing to find something obviously fishy but that gets through the algorithm for some obscure reason.
Sometimes simple approaches are the way to go. You're going to have false positives and false negatives no matter what, the question is how much and in what circumstances. And this particularly scenario is unlikely to be all that common.
Tweak your mailer so that it sends mail from gi-request instead of google-issues-request, and don't mention "Google Account". Granted, this sucks, but the Internet routes around brokenness, and that's what you need to do in a situation like this. Is that a sad thing? Yes, of course. If we had a mail architecture that was pull- rather than push-based, maybe we could have nice things, but until that magic day, the whole thing is bubble gum and bailing wire, and it's honestly not Google's fault that that's so.
As another example of brokenness, I often get mail that is marked spam because it went through a mailing list expander and the headers didn't get rewritten, so that it fails DKIM validation. Yes, we can all rail about how evil and awful DKIM is, but the bottom line is that if you don't want that to happen, you rewrite the headers. Again, a system that's pull-based rather than push-based would make this a lot better.
C'mon, Lauren, with the 10's of millions of spams that google catches every day, some things are going to get caught by the filter that shouldn't be. Even if the filter is 99.99 effective that means there will be 1000 false positives in there...and yours is one of them. Shit happens. Adjust and move on.
Apparently what we're dealing with here is a simplistic (and frankly, rather haphazard in this respect at least) string-matching algorithm that could have come right out of the early 1970s...! [A]t least in this case, it appears that Google is basically using the venerable old UNIX/Linux "grep" command or some equivalent, and in a rather slipshod way, too. is drawing a trend and a conclusion from one data point.
GMail won't normally mark your email as spam/phishing if you've implemented basic mail server identification such as SPF (Sender Policy Framework) and DKIM (DomainKeys Identified Mail). This is well known, and I guarantee that if the author bothered to search for why their mail ends up flagged by GMail he would hit at least one of these two terms in the first few results.
A recursive sig
Can impart wisdom and truth
Call proc signature()
You mean like that vortex.com front page?
Beware of the Leopard.
With the huge volumes of data that Google handles, it's probably hard to do any better.
GMail may be "hard to do any better", but dealing with spam is complex and labyrinthine.
My client began having conversations with a vendor last week, and as a result GMail put *all* subsequent E-mails into my spam folder, including ones from my (whitelisted) client addressed to the vendor CC'ing me. I only found out by accident.
One might *expect* a quick, easily identified control that says "whitelist this person" or "whitelist this company", but there isn't. You have to go to "Settings->Settings->Filters and blocked addresses", none of which terms are "spam", so the casual user can't just scan headings for the term.
You can't, apparently, just refer to the spam and say "whitelist that person", you need to create a new filter. You can't, apparently, say "@example.com" as a wildcard for the business, you have to identify an actual sender by complete address.
And of course, you have to discover that you need to do this, because GMail doesn't give any warning. (Surprising, since every time I use GMail from a different location it sends me a warning E-mail. Every. Single. Time.)
I'm not even sure why everything went to spam in the first place - I had sent E-mails to both the vendor and the client, so they should have been in my "recently used" list.
GMail has a pretty cryptic interface, compared to some of the other mail readers I've used.