[HN Gopher] Google Safe Browsing missed 84% of confirmed phishin...
       ___________________________________________________________________
        
       Google Safe Browsing missed 84% of confirmed phishing sites
        
       Author : jdup7
       Score  : 262 points
       Date   : 2026-03-05 15:03 UTC (7 hours ago)
        
 (HTM) web link (www.norn-labs.com)
 (TXT) w3m dump (www.norn-labs.com)
        
       | supermatt wrote:
       | > When we ran the full dataset through the deep scan, it caught
       | every single confirmed phishing site with zero false negatives.
       | The tradeoff is that it flagged all 9 of the legitimate sites in
       | our dataset as suspicious
       | 
       | Huh? Does this mean it just flagged everything as suspicious?
        
         | badgersnake wrote:
         | lol, return false;
        
         | john_strinlai wrote:
         | indeed... it seems like it just says everything is phishing...
         | which they go on to say is desirable?
         | 
         | "The tradeoff is that it flagged all 9 of the legitimate sites
         | in our dataset as suspicious, _which is worth it when you 're
         | actively investigating a link you don't trust._"
         | 
         | so, you dont really need the scanning product at all. if you
         | just assume every website is a phishing website, you will have
         | the same performance as the scanner!
        
         | jdup7 wrote:
         | Yeah probably could have done better at describing the
         | methodology. The dataset is just the confirmed (manually by a
         | human) phishing urls. We only included the FPs to show that the
         | tooling isn't perfect there were many TNs that we did not
         | include. Going forward we could definitely frame these results
         | better.
        
       | lich_king wrote:
       | I don't understand the metric they're using. Which is maybe to be
       | expected of an article that looks LLM-written. But they started
       | with ~250 URLs; that's a _weirdly small_ sample. I 'm sure there
       | are tens of thousands malicious websites cropping up monthly. And
       | I bet that Safe Browsing flags more than 16% of that?
       | 
       | So how did they narrow it down to that small number? Why these
       | sites specifically?... what's the false positive / negative rate
       | of both approaches? What's even going on?
        
         | jdup7 wrote:
         | Probably could have been a bit more descriptive around the
         | dataset. Our tooling pulls in a lot more than 250 URLs but
         | since we are manually confirming them that means a smaller
         | dataset. In other words, out of the urls we pulled in these 250
         | were confirmed (by a human) as phishing. We did not do any
         | selection beyond that. As for the article LLMs were used to
         | help with the graphs and grammatical checks but that's it. This
         | was our first month of going through this exercise and we
         | definitely want to have larger datasets going forward as we
         | expand capacity for review.
         | 
         | As for Safe Browsing catching more than 16% it depends on the
         | timeline at the time these attacks are launched it's likely
         | Safe Browsing catches closer to 0% but as the time goes on that
         | number definitely climbs.
        
         | john_strinlai wrote:
         | > _what 's the false positive / negative rate of both
         | approaches_
         | 
         | the false positive rate is 100%. they just say everything is
         | phishing:
         | 
         | " _When we ran the full dataset through the deep scan, it
         | caught every single confirmed phishing site with zero false
         | negatives. The tradeoff is that it flagged all 9 of the
         | legitimate sites in our dataset as suspicious, which is worth
         | it when you 're actively investigating a link you don't
         | trust._"
        
           | lorenzoguerra wrote:
           | it's 100% for what they call "deep scan", it's 66.7% for the
           | "automatic scan". Practically unusable anyway
        
       | PunchyHamster wrote:
       | They put them directly in front of search results, why would they
       | not miss them ?
        
       | xvector wrote:
       | There's probably like one engineer maintaining this as a side
       | project at the company
        
         | andor wrote:
         | Yeah, it would be interesting to know how much work is spent on
         | it. I sometimes submit sites when I am targeted by a campaign,
         | but I'm not sure if they end up in their deny-list.
        
       | candiddevmike wrote:
       | I'm getting some kind of chrome security warning when using
       | zscaler now. Discussing all of this with non-techies, I think
       | folks are overwhelmed by all of the security warnings they get
       | and have stopped paying attention to them.
       | 
       | So what's the point of doing all of this if there isn't some kind
       | of corresponding education on responsible computer use? There
       | needs to be some personal responsibility here, you can't protect
       | people against everything.
        
       | dvh wrote:
       | Just yesterday I marked another Gmail phishing scam. This
       | wouldn't be worth mentioning but they are using Google's own
       | service for it. It has to be intentional, there is no other
       | explanation. https://news.ycombinator.com/item?id=46665414
        
       | iqandjoke wrote:
       | But why Apple choose to work with this on Safari?
        
       | nico wrote:
       | On a tangent - gmail has a feature to report phishing emails, but
       | it seems like it's only available on the website. Their mobile
       | app doesn't seem to have the option (same with "mark as unread").
       | Is it hidden or just not available?
        
         | bradyd wrote:
         | The mobile app definitely has mark as unread. It's the envelope
         | icon next to the trashcan (the exact same icon as in the web
         | interface). Never realized there was a report phishing option.
         | I just mark those emails as spam, which is available in the
         | app.
        
       | itvision wrote:
       | Criminals can easily show Google crawlers "good" websites.
       | 
       | The fact that Safe Browsing even works is already good enough.
        
       | 7777777phil wrote:
       | Blocklists assume you can separate malicious infrastructure from
       | legitimate infrastructure. Once phishing moves to Google Sites
       | and Weebly that model just doesn't work.
        
       | lorenzoguerra wrote:
       | >We also ran the full dataset of 263 URLs (254 phishing, 9
       | confirmed legitimate) through Muninn's automatic scan. This is
       | the scan that runs on every page you visit without any action on
       | your part. On its own, the automatic scan correctly identified
       | 238 of the 254 phishing sites and only incorrectly flagged 6
       | legitimate pages.
       | 
       | ...so it has a false positive rate of 67%? On a ridiculously
       | small dataset?
        
         | jdup7 wrote:
         | Fair point in isolation that number doesn't look good. The
         | important context is that this dataset was built to test
         | phishing detection, not to measure false positive rates on
         | normal traffic. It's sourced from our threat intelligence
         | tooling so it's almost entirely malicious URLs by design. The 9
         | clean sites aren't a random sample of everyday browsing.
         | They're sites that were submitted as suspicious and turned out
         | to be legitimate so they're basically the hardest possible set
         | of clean pages to correctly classify. This seems like a common
         | critique and we definitely could have done a better job of
         | explaining the methodology. Going forward we will include
         | numbers from daily use to give a better picture of FP rate.
        
       | mholt wrote:
       | I never loved the idea of GSB or centralized blocklists in
       | general due to the consequences of being wrong, or the
       | implications for censorship.
       | 
       | So for my masters' thesis about 6-7 years ago now (sheesh) I
       | proposed some alternative, privacy-preserving methods to help
       | keep users safe with their web browsers:
       | https://scholarsarchive.byu.edu/etd/7403/
       | 
       | I think Chrome adopted one or two of the ideas. Nowadays the
       | methods might need to be updated especially in a world of LLMs,
       | but regardless, my hope was/is that the industry will refine some
       | of these approaches and ship them.
        
         | notepad0x90 wrote:
         | Block lists will always be used for one reason or another, in
         | this case these are verified malicious sites, there is no
         | subjective analysis element in the equation that could be
         | misconstrued as censorship. But even if there was, censorship
         | implies a right to speech, in this case Google has the right to
         | restrict the speech of it's users if it so wishes, matter of
         | fact, through extensions there are many that do censor their
         | users using Chrome.
        
           | rstupek wrote:
           | I know for a fact that GSB contains non-malicious sites in
           | its dataset.
        
       | sirpilade wrote:
       | But hits 100% of browsing tracking
        
         | blell wrote:
         | Educate yourself on how it works before you say something like
         | this.
        
           | sirpilade wrote:
           | Pun aside, I cannot fully trust a centralized URL checker on
           | a remote server that I don't own, even if they guarantee that
           | my privacy is safe
        
       | notepad0x90 wrote:
       | Glass is half empty, I see.
       | 
       | How about GSB stopped 16% of phishing sites? that's still huge.
        
         | debo_ wrote:
         | I guess the glass is 16% full.
        
         | loloquwowndueo wrote:
         | Would you use anything that was only 16% effective for its
         | claimed purpose?
         | 
         | "Tylenol stops headaches in 16% of people" - it's huge, right?
         | That's millions of people we're talking about.
         | 
         | Would you use it?
        
           | mock-possum wrote:
           | Idk why not? What're the side effects?
        
           | HeatrayEnjoyer wrote:
           | Countless medications have <16% efficacy rate.
        
           | NekkoDroid wrote:
           | If the other options would just straight up kill innocent
           | bystanders (e.g. false positives for legit shops) I think
           | that is a tradeoff I am willing to make.
        
           | asadotzler wrote:
           | 99% of users don't even know they're being protected. There's
           | no promise except "we work to make browsing safer" and
           | cutting even 5% of malicious sites from a user's experience
           | is an unmitigated win for that user at the low false positive
           | rate Safe Browsing offers.
        
       | epicprogrammer wrote:
       | Having spent some time in the anti-abuse and Trust & Safety
       | space, I always take these vendor reports with a massive grain of
       | salt. It's a classic case of comparing apples to vendor-marketing
       | oranges. A headline screaming about an 84% miss rate sounds like
       | a systemic collapse until you look at the radically different
       | constraint envelopes a global default like GSB and a specialized
       | enterprise vendor operate under.
       | 
       | The biggest factor here is the false-positive cliff. Google Safe
       | Browsing is the default safety net for billions of clients across
       | Chrome, Safari, and Firefox. If GSB's false-positive rate ticks
       | up by even a fraction of a percent, they end up accidentally
       | nuking legitimate small businesses, SaaS platforms, or municipal
       | portals off the internet. Because of that massive blast radius,
       | GSB fundamentally has to be deeply conservative. A boutique
       | security vendor, on the other hand, can afford to be highly
       | aggressive because an over-block in a corporate environment just
       | results in a routine IT support ticket.
       | 
       | You also have to factor in the ephemeral nature of modern
       | phishing infrastructure and basic selection bias. Threat actors
       | heavily rely on automated DGAs and compromised hosts where the
       | time-to-live for a payload is measured in hours, if not minutes.
       | If a specialized vendor detects a zero-day phishing link at 10:00
       | AM, and GSB hasn't confidently propagated a global block to
       | billions of edge clients by 10:15 AM, the vendor scores it as a
       | "miss." Add in the fact that vendors naturally test against the
       | specific subset of threats their proprietary engines are tuned to
       | find, and that 84% number starts to make a lot more sense as a
       | top-of-funnel marketing metric rather than a scientific baseline.
       | 
       | None of this is to say GSB is perfect right now. It has
       | absolutely struggled to keep up with the recent explosion of
       | automated, highly targeted spear-phishing and MFA-bypass proxy
       | kits. But we should read this report for what it really is: a
       | smart marketing push by a security vendor trying to sell a
       | product, not a sign that the internet's baseline immune system is
       | totally broken.
        
         | Medowar wrote:
         | > We also ran the full dataset of 263 URLs (254 phishing, 9
         | confirmed legitimate) through Muninn's automatic scan. This is
         | the scan that runs on every page you visit without any action
         | on your part. On its own, the automatic scan correctly
         | identified 238 of the 254 phishing sites and only incorrectly
         | flagged 6 legitimate pages. [...] The tradeoff is that it
         | flagged all 9 of the legitimate sites in our dataset as
         | suspicious, ...
         | 
         | Am I missing something or is that a 66%/100% False Positive
         | Rate on legitimate Sites?
         | 
         | If GSB would have that ratio, it would be absolute unusable..
         | So comparing these two is absolutely wrong...
        
           | ApolloFortyNine wrote:
           | The 9/9 is actually crazy, and then they posted about it as
           | if they found something? What they did was find a major issue
           | in their own process and then told the world about it, that
           | just doesn't seem right.
        
             | trehalose wrote:
             | It would seem their service identifies _only_ phishing
             | sites as legitimate ones. It would seem 100% of sites they
             | deem legitimate are phishing sites. Incredible.
        
             | thrwaway55 wrote:
             | The deep scan detected all phishing sites correctly with
             | the unfortunate tagging of legit sites as phishing too. I
             | imagine their code looks something like isPhishing = true.
        
               | andoando wrote:
               | lol
        
             | saalweachter wrote:
             | Crazy, and also like, 9? The sample size in that part of
             | your test suite is _9_?
        
         | ajross wrote:
         | > I always take these vendor reports with a massive grain of
         | salt.
         | 
         | Yeah. "Here's a blog post with some casually collected numbers
         | about our product [...] It turns out that it's great!" is sorta
         | boring.
         | 
         | But couple that with a headline framed as "Google [...] Bad"
         | and straight to the top of the HN front page it goes!
        
         | jdup7 wrote:
         | These are fair points and I agree with a lot of them. GSB
         | operates at a scale we don't, and the conservatism that comes
         | with being the default for billions of users is a real
         | constraint. The post tries to acknowledge that ("the takeaway
         | from all of this is not that Google Safe Browsing is bad") and
         | we're upfront about the timing caveat since these were checked
         | at time of scan.
         | 
         | Where I'd push back is on what this means for the average
         | person. Most people have no protection against phishing beyond
         | what their email provider and browser give them. If that
         | protection is fundamentally reactive, catching threats hours or
         | days after they go live, that's a real limitation worth talking
         | about honestly. The 84% number isn't meant to say GSB is
         | broken. It's meant to say there's a gap, and that gap has
         | consequences for real users regardless of the engineering
         | reasons behind it.
         | 
         | On the marketing angle, we aren't currently selling anything.
         | The extension is free and so is submitting URLs for
         | verification. We recognize it would be disingenuous to say we
         | never will, but at the very least the data and the ability to
         | check URLs (similar to PhishTank before they closed
         | registration) will always be free. The dataset is also sourced
         | from public threat intelligence feeds, not a curated set
         | designed to make our tool look good. We think publishing
         | findings like this is valuable even if you set aside everything
         | about our tools.
        
           | philipallstar wrote:
           | > We think publishing findings like this is valuable even if
           | you set aside everything about our tools.
           | 
           | In what way is it valuable?
        
         | nubinetwork wrote:
         | > I always take these vendor reports with a massive grain of
         | salt. It's a classic case of comparing apples to vendor-
         | marketing oranges. A headline screaming about an 84% miss rate
         | sounds like a systemic collapse until...
         | 
         | I've seen this before in the ip blocklist space... if you're
         | layering up firewall rules, you're bound to see the higher
         | priority layers more often.
         | 
         | That doesn't mean the other layers suck, security isn't always
         | an A or B situation...
         | 
         | On the other hand, I don't know how I feel about how GSB is
         | implemented... you're telling google every website you go to,
         | but chances are the site already has google analytics or SSO...
        
           | throawayonthe wrote:
           | i thought it was checks against a local list of hashes? with
           | frequent updates
        
             | 7v3x3n3sem9vv wrote:
             | this is how Firefox does it. can't speak for the rest.
        
       | xnx wrote:
       | Why should I trust that "Norn Labs" knows what is and is not a
       | phishing site?
        
       | mrexcess wrote:
       | These statistics would be a lot better if they were compared
       | directly to the same measurements taken from dedicated cloud
       | SWGs/SSEs like Zscaler. My somewhat subjective sense is that the
       | whole industry is in a bit of a rough patch, the miss rate seems
       | to be noticeably climbing all across the board.
        
       | pothamk wrote:
       | One thing that often gets overlooked in these comparisons is
       | distribution latency.
       | 
       | Detecting a phishing domain internally is one problem, but
       | pushing a verified block to billions of browsers worldwide is a
       | completely different operational challenge.
       | 
       | Systems like Safe Browsing have to worry about propagation time,
       | cache layers, update intervals, and the risk of pushing a false
       | positive globally. A specialized vendor can update instantly for
       | a much smaller customer base.
       | 
       | That difference alone can easily look like a "miss" in snapshot-
       | style measurements.
        
         | tadfisher wrote:
         | If you are not a bot, I suggest changing your voice so that you
         | are distinguishable from one. You're not _wrong_ , just like
         | you weren't wrong about "one thing that trips people up about
         | asyncio" yesterday, but I noticed the slop-speak immediately.
         | I'm sure others have as well.
        
           | niwtsol wrote:
           | Multiple comments that start w/ "what's interesting about" by
           | this user and very similar formatting kind of answers that
           | question on human vs bot. Weird internet we live in these
           | days.
        
       | timnetworks wrote:
       | The most dangerous links recently have been from sharepoint.com,
       | dropbox.com, etc. and nobody is going to block those.
        
       | varispeed wrote:
       | When Google will remove scams, phishing and other nonsense from
       | their advertising? Especially the scareware stuff, where AI
       | videos say someone might be listened to / hacked and here is the
       | software that will help block it / find it whatnot. Then they
       | collect personal data.
        
       | passwordoops wrote:
       | Anecdotal and loosely related, but I can say since Gemini was
       | forced into Gmail, much more obvious SPAM passes the filter
        
       | hedora wrote:
       | So, the false negative rate was 84%, but what was the false
       | positive rate?
       | 
       | They have a table "AUTOMATIC SCAN RESULTS (263 URLS)" that sort
       | of presents this information. Of the 9 sites that were negatives,
       | they say they incorrectly flagged 6 as phishing.
       | 
       | With a false positive rate of 66%, it's not surprising they were
       | able to drive down their false negative rate. Also, the test set
       | of 254 phishing sites with 9 legitimate ones is a strange choice.
       | 
       | (Or maybe they need to work on how they present data in tables;
       | tl;dr the supporting text.)
        
         | decimalenough wrote:
         | The false positive rate was 66% for "automatic scan" and 100%
         | (!) for "deep scan".
         | 
         | In other words, you can get these numbers if your deep scan
         | filter is isSuspicious() { return true; }.
        
           | antonvs wrote:
           | Brb, applying for YC funding for my new AI-based phishing
           | detection system.
           | 
           | ('return true' is just a very optimized neural network after
           | all!)
        
       | obblekk wrote:
       | Maybe I'm an outlier but I'd rather this than accidentally block
       | legit sites.
       | 
       | Otherwise this becomes just another tool for Google to wall in
       | the subset of the internet they like.
        
       | caaqil wrote:
       | Yeah, maybe let's change the title to remove that 84% rate. It's
       | meaningless because it's just 254 websites, given the scale of
       | what Google Safe Browsing deals with.
       | 
       | How is this serious? This is a marketing slop. If the title isn't
       | enough indicator, the ending should be:
       | 
       | > _If you 're interested in trying Muninn, it's available as a
       | Chrome extension. We're in an early phase and would genuinely
       | appreciate feedback from anyone willing to give it a shot. And if
       | you run across phishing in the wild, consider submitting it to
       | Yggdrasil so the data can help protect others._
        
       | bethekidyouwant wrote:
       | Their example is really dumb. Eventually, you get a fake
       | Microsoft login page, but they clip out the address bar which
       | clearly isn't a Microsoft address so your auto complete password
       | isn't going to be put into the form and you'd have to be pretty
       | dumb to type it in my hand or even to know your Microsoft
       | password, it should be some random thing generated by Safari or
       | whatever your password manager is. Not to mention two factor
       | authentication.
        
       | dsr_ wrote:
       | Almost all email phishing attempts we receive come from GMail.
        
       | kemotep wrote:
       | Default deny and only permitting what you explicitly allow stops
       | 90% of this in a corporate environment.
       | 
       | You don't just leave all your ports open on the firewall and only
       | close the ones exploited. You default deny and only allow the
       | bare minimum you need in.
        
       | virken wrote:
       | I'm all for stopping phishing - and the tool sounds great - but I
       | have to say the Web Store Extension listing is very concerning -
       | even with a new company/offering - there's only 4 users - and 1
       | rating (a 5 of course) - I'd like to try - but seems phishy :-(
        
       | virken wrote:
       | I'm all for stopping phishing - and the tool sounds great - but
       | the Web Store Extension listing is very concerning - even with a
       | new company/offering - there's only 4 users - and 1 rating (a 5
       | of course) - I'd like to try - but seems phishy :-(
        
       | thayne wrote:
       | It would be interesting to see how many of the sites safe
       | browsing does block are false positives.
        
       | nickphx wrote:
       | "If you're interested in trying Muninn, it's available as a
       | Chrome extension. We're in an early phase " Domain is less than 4
       | months old.. Software is "early phase".. Already making
       | misleading marketing claims of usefulness..
        
       | kopollo wrote:
       | Let me give you a simple detection algorithm. Apply OCR to the
       | screenshot because they often use logos. Also, parse the text
       | from the HTML and compare it to the URL. You can catch a lot of
       | spam this way.You can also examine many parameters in the js html
       | code.
        
       | throawayonthe wrote:
       | > ...full dataset of 263 URLs (254 phishing, 9 confirmed
       | legitimate)
       | 
       | > ... automatic scan is optimized for precision (keeping false
       | alarms low...
       | 
       | really?
       | 
       | > When we ran the full dataset through the deep scan, ... it
       | flagged all 9 of the legitimate sites in our dataset as
       | suspicious
       | 
       | lol
        
       | hulitu wrote:
       | The purpose of "Safe Browsing" is to send your URLs to Google.
        
       | nickphx wrote:
       | So I tested out the extension.. First the extension spammed me
       | with "login required".. So I click the notification to be taken
       | to a login page.. Great? Now I have to create an account and
       | verify a link.. Now I can test how great this is against a
       | "fresh" facebook phishing page being actively promoted via
       | Facebook Ads..
       | 
       | hxxps://r7ouhcqzdgae76-fsc0fydmbecefrap.z03.azurefd.net/new2/?utm
       | _medium=paid&utm_source=fb&utm_id=6900429311725&utm_content=69004
       | 29312725&utm_t erm=6900429314125&utm_campaign=6900429311725
       | 
       | The "extension" did a "scan". {"url":"https://r7ouhcqzdgae76-fsc0
       | fydmbecefrap.z03.azurefd.net/new2..."}
       | 
       | response: {"classification":"clean"}
       | 
       | great work?
       | 
       | If I click "Deep scan".. I see a screenshot blob being sent
       | over.. response: { "classification": "phish", "reasons": [ "Our
       | system has previously flagged this webpage as malicious." ] }
       | 
       | So if the site were already flagged, why does the "light" scan
       | not show that?
        
       | luckman212 wrote:
       | Is it possible to disable Safe Browsing AND also _not_ have to
       | manually click to confirm that _" yes, I actually do want to keep
       | the file I just downloaded, thank you"_ every. single. time.
        
         | burnte wrote:
         | No. I keep checking every year.
        
       ___________________________________________________________________
       (page generated 2026-03-05 23:01 UTC)