Thousands of systems, from Google to script kiddies to OpenAI to nigerian call scammers to cybersecurity firms, actively watch the certificate transparency logs for exactly this reason. Yawn.
Considering how it must be getting hammered what with the "AI" nonsense, it's interesting how crt.sh continues to remain usable, particularly the (limited) direct PostgresSQL db access
To me, this is evidence that SQL databases with high traffic can be made directly accessible on the public internet
crt.sh seems to be more accessible at certain times of the day. I can remember when it had no such accessibility issues
It's the only website I know of where queries can just randomly fail for no reason, and they don't even have an automatic retry mechanism. Even the worst enterprise nightmares I've seen weren't this user unfriendly.
With that said, given that (1) pre-certificates in the log are big and (2) lifetimes are shortening and so there will be a lot of duplicates, it seems like it would be good for someone to make a feed that was just new domain names.
(It doesn't deduplicate if the same domain name appears in multiple certificates, but it's still a substantial reduction in bandwidth compared to serving the entire (pre)certificate.)
EDIT: that's the flip side of supporting HTTPS that's not well-known among developers - by acquiring a legitimate certificate for your service to enable HTTPS, you also announce to the entire world, through a public log, that your service exists.
I don't really see how this is a flip-side. If you're putting something on the web, presumably you want it to be accessed by others, so this is actually a benefit.
If you didn't want others to access your service, maybe consider putting it in a private space.
If certificate transparency is new to you, I feel like there are significantly more interesting articles and conversations that could/should have been submitted instead of "A public log intended for consumption exists, and a company is consuming that log". This post would do literally nothing to enlighten you about CT logs.
If the fact that OpenAI is scraping certificate transparency logs is new and interesting to you, I'd love to know why it is interesting. Perhaps I'm missing something.
Way more interesting reads for people unfamiliar with what certificate transparency is, in my opinion, than this "OpenAI read my CT log" post:
> I feel like there are significantly more interesting articles
if this is the article that introduces someone to the concept of certificate transparency, then there's nothing wrong with that. graciously, you followed through with links to what you consider more interesting. that is not something a lot of commenters do and just leave it as a snarky comment for someone being one of the lucky 10000 for the day.
Yeah, this is the unspoken part about HTTPS: you enable it, you also announce to the entire world you're serving stuff from specific DNS names :).
(Which is why I hate it that it's so hard to test things locally without having to get a domain and a certificate. I don't want to buy domain names and announce them publicly for the sake of some random script that needs to offer a HTTP endpoint.)
Modern security is introducing a lot of unexpected couplings into software systems, including coupling to political, social and physical reality, which is surprising if you think in terms of programs you write, which most likely shouldn't have any such relationships.
My favorite example of such unexpected coupling, whose failures are still regularly experienced by users, is wall clock time. If your program touches anything related to certificates, even indirectly, suddenly it's coupled to actual real clock and your users better make sure their system time is in synch with the rest of the world, or else things will stop working.
Even if it's just for their internal security initiatives it would make sense given how massive they are. Threat hunting via cert monitoring is very effective.
Presumably this is well-known among people that already know about this.
P.S. In the hopes of making this more than just a sarcastic comment, the question of "How do people bootstrap knowledge?" is kind of interesting. [1]
> To tackle a hard problem, it is often wise to reuse and recombine existing knowledge. Such an ability to bootstrap enables us to grow rich mental concepts despite limited cognitive resources. Here we present a computational model of conceptual bootstrapping. This model uses a dynamic conceptual repertoire that can cache and later reuse elements of earlier insights in principled ways, modelling learning as a series of compositional generalizations. This model predicts systematically different learned concepts when the same evidence is processed in different orders, without any extra assumptions about previous beliefs or background knowledge. Across four behavioural experiments (total n = 570), we demonstrate strong curriculum-order and conceptual garden-pathing effects that closely resemble our model predictions and differ from those of alternative accounts. Taken together, this work offers a computational account of how past experiences shape future conceptual discoveries and showcases the importance of curriculum design in human inductive concept inferences.
The whole point of the CT logs is to be a public list of all domains which have TLS certs issued by the Web PKI. People are reading this list. I really don't see what is either surprising or in any way problematic in doing so.
This could be OpenAI, or it could be another company using their header pattern.
It has long been common for scrapers to adopt the header patterns of search engine crawlers to hide in logs and bypass simple filters. The logical next step is for smaller AI players to present themselves as the largest players in the space.
Some search engines provide a list of their scraper IP ranges specifically so you can verify if scraper activity is really them or an imitator.
EDIT: Thanks to the comment below for looking this up and confirming this IP matches OpenAI’s range.
I don't know if imitating a major crawler is really worth it, it may work against very naive filters, but it's easy to definitively check whether you're faking so it's just handing ammo to more advanced filters which do check.
I don't have a statistic here, but I'm always surprised how many websites I come across that do limited user-agent and origin/referrer checks, but don't maintain any kind of active IP based tracking. If you're trying to build a site-specific scraper and are getting blocked, mimicking headers is an easy and often sufficient step.
> Some search engines provide a list of their scraper IP ranges
Common Crawl's CCBot has published IP ranges. We aren't a search engine (although there are search engines using our data) and we like to describe our crawler as a crawler, not a "scraper".
I don't understand the outrage in some of the comments. The certificate transparency logs are literally meant to be read by absolutely whoever wants to read them. The clue is right in the name. It's transparency logs! Transparency!
I just don't understand how people with no clue whatsoever about what's going on feel so confident to express outrage over something they don't even understand! I don't mind someone not knowing something. Everybody has to learn things somewhere for the first time. But couldn't they just keep their outrage to themselves and take some time to educate themselves, to find out whether that outrage is actually well placed?
Some of the comments in the OP are also misinformed or illogical. But there's one guy there correcting them so that's good. I mean I'd say that https://en.wikipedia.org/wiki/Certificate_Transparency or literally any other post about CT is going to be far more informative than this OP!
People are just raging and want an outlet. They aren’t thinking logically.
It’s been going on forever (remember how companies were reading files off your computer aka cookies in 1999?)
This seems like a total non-issue and expect that any public files are scraped by OpenAI and tons others. If I don’t want something scraped, I don’t make it public.
Yep, but this comes with a tradeoff: all of your services now have a valid key/cert for your whole domain, significantly increasing the blast radius if one service is compromised.
This is what we do for development containers/hosts - put them behind *.dev.example.com, allows us to hide most testing instances using a shared load balancer. And with a single wildcard CNAME, No info is leaked in CT logs or DNS. Said LB is firewalled, but why pay for extra traffic that's just going to be blocked?
May I know does Caddy automatically update with apt if you built custom Caddy binaries for the DNS provider plugin?
Also, may I know which DNS provider you went with? The GitHub issues pages with some of the DNS provider plugins seems to suggest some are more frequently maintained, while some less so.
Unfortunately they are a bit extra bothersome to automate (depending on your DNS provider/setup) because of the DNS CNAME-method validation requirement.
Yep, but next year they intend to launch an alternative DNS challenge which doesn't require changing DNS records with every renewal. Instead you'll create a persistent TXT record containing a public key, and then any ACME client which has the private key can keep requesting new certs forever.
Oh, sweet! I didn't know about this. I have no need of wildcard certs, but this will greatly simplify the process of issuing certificates for internal services behind my local firewall. No need to maintain an acme-dns server; just configure the ACME client, set the DNS record and you're done? Very nice.
If you are using a non-standard DNS provider that doesn’t have integration with certbot or cert-manager or whatever you are using, it is pretty easy to set up an acme-dns server to handle it
I hadn't heard of Agnos before, interesting alternative to ACME-DNS.
Looking at the README, is the idea that the certificates get generated on the DNS server itself? Not by the ACME client on each machine that needs a certificate? That seems like a confusing design choice to me. How do you get the certificate back to the web server that actually needs it? Or is the idea that you'd have a single server which acts as both the DNS server and the web server?
For many years now. The crawlers, scanners, and bots start hammering a website within a minute of a certificate being issued. Remember to get your garbage WCM installed and secured before installing the real certificate as you have about a 15 second window before they're hammering around for fresh wordpress installs. Granted, you people are all smart enough to have all that automated using a CI/CD pipeline so that you just commit a single file with the domain name to a git repo and all that magic happens.
The ostensible purpose of the certificate transparency logs is to allow validation of a certificate you're looking at - I browse to https://poormathskills.com and want to figure out the details of when its cert was issued.
The (presumably) unintended, unexpected purpose of the logs is to provide public notification of a website coming online for scrapers, search engines, and script kiddies to attack it: I could register https://verylongrandomdomainnameyoucantguess7184058382940052... and unwisely expect it to be unguessable, but as it turns out OpenAI is going to scrape it seconds after the certificate is issued.
Yet they provide the user agents and IP address ranges which they scrape from, and say they respect robots.txt
I run a web server and so see a lot of scrapers, but OpenAI is one of the ones that appear to respect limits that you set. A lot of (if not most) others don't even have that ethics standard so I'd not say that "OpenAI scrapes everything they can access. Everything" without qualification, as that doesn't seem to be true, at least not until someone puts a file behind a robots deny page and finds that chatgpt (or another of openai's products) has knowledge of it
There's no evidence the barrage of residentially-proxied bot accesses hitting every public website have anything to do with OpenAI, but then again, there's also no evidence they don't.
If you want to learn more about Certificate Transparency Logs, how to pull and search them, we just did a 3 part series about how we did this at CertKit:
https://www.certkit.io/blog/searching-ct-logs
Seems like you could set up a cert for a honeypot domain to collect ips of bots running off of the certificate transparency logs. If domain isnt linked from anywhere, then its pretty sure to be a bot isn't it?
Given these are trivially forged, presumably they aren't really using a Mac for scraping, right? Just to elicit a 'standard' end user response from the server?
>useragent="Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36; compatible; OAI-SearchBot/1.3; robots.txt;
Makes me want to reconfigure my servers to just drop such traffic. If you can't be arsed to send a well-formed UA, I have doubts that you'll obey other conventions like robots.txt.
They definitely do. Before this comment CT logs – aside from DNS queries – were the only way to know about https://onion.basilikum.monster and you have to send the hostname in the SNI, otherwise you get another certificate back.
Of course I get some bot traffic including the OpenAI bot – although I just trust the user agent there, I have not confirmed that the origin IP address of the requests actually belongs to OpenAI.
That's just the internet. I like the bots, especially the wp-admin or .env ones. It's fun watching them doing their thing like little ants.
So? It’s public information and a somewhat easily consumable stream of websites to scrape, if my job was to scrape the entire internet I’d probably start there, too.
But what GP probably meant is that OAI definitely uses this log to get a list of new websites in order to scrap then later. This is a pretty standard way to use CT logs - you get a list of domains to scrap instead of relying solely on hyperlinks.
I think their point is that the people registering certs may not intend their sites to be immediately scraped, but now OpenAI is bypassing e.g. google indexing or web spidering, and using your cert provider's CT entries to find you immediately for scraping.
Looking around at the comments, I have a birds-eye view. People are quite skilled at jumping to conclusions or assuming their POV is the only one. Consider this simplified scenario to illustrate:
- X happened
- Person P says "Ah, X happened."
- Person Q interprets this in a particular way
and says "Stop saying X is BAD!"
- Person R, who already knows about X...
(and indifferent to what others notice
or might know or be interested in)
...says "(yawn)".
- Person S narrowly looks at Person R and says
"Oh, so you think Repugnant-X is ok?"
What a train wreck. Such failure modes are incredibly common. And preventable.* What a waste of the collective hours of attention and thinking we are spending here that we could be using somewhere else.
See also: the difference between positive and normative; charitable interpretations; not jumping to conclusions; not yucking someone else's yum
* So preventable that I am questioning the wisdom of spending time with any communication technology that doesn't actively address these failures. There is no point at blaming individuals when such failures are a near statistical certainty.
I agree with your analysis but try to not agree with your conclusion, purely for my own metal hygiene: I believe one can retrain the pattern matching of one’s brain for happier outcomes. If I let my brain judge this as a “failure“ (judgment “it is wrong“), I will either get sad about it (judgment “… and I can’t change it“) or angry (… and I can do something about it“). In cases such as this I prefer to accept it as is, so I try to rewrite my brain rule to consider it a necessary part of life (judgment “true/good/correct“).
Ah, in case it didn't come across clearly, my conclusion isn't to blame the individuals. My assessment is to seek out better communication patterns, which is partly about "technology" and partly about culture (expectations). People could indeed learn not to act this way with a bit of subtle nudging, feedback, and mechanism design.
I'm also pretty content accepting the unpleasant parts of reality without spin or optimism. Sometimes the better choice is still crappy, after all ;) I think Oliver Burkeman makes a fun and thoughtful case in "The Antidote: Happiness for People Who Can't Stand Positive Thinking" https://www.goodreads.com/book/show/13721709-the-antidote
(the site may occasionally fail to load)
https://www.merklemap.com/search?query=ycombinator.com&page=...
Entries are indexed by subdomain instead of by certificate (click an entry to see all certificates for that subdomain).
Also, you can search for any substring (that was quite the journey to implement so it's fast enough across almost 5B entries):
https://www.merklemap.com/search?query=ycombi&page=0
[0] https://github.com/Barre/ZeroFS
Love it
To me, this is evidence that SQL databases with high traffic can be made directly accessible on the public internet
crt.sh seems to be more accessible at certain times of the day. I can remember when it had no such accessibility issues
It's the only website I know of where queries can just randomly fail for no reason, and they don't even have an automatic retry mechanism. Even the worst enterprise nightmares I've seen weren't this user unfriendly.
For example:
(It doesn't deduplicate if the same domain name appears in multiple certificates, but it's still a substantial reduction in bandwidth compared to serving the entire (pre)certificate.)Needs clarification. What reason
EDIT: that's the flip side of supporting HTTPS that's not well-known among developers - by acquiring a legitimate certificate for your service to enable HTTPS, you also announce to the entire world, through a public log, that your service exists.
If you didn't want others to access your service, maybe consider putting it in a private space.
Certificate transparency logs are intended to be consumed by others. That is indeed what is happening. Not interesting.
It's certainly news to me, and presumably some others, that this exists.
If certificate transparency is new to you, I feel like there are significantly more interesting articles and conversations that could/should have been submitted instead of "A public log intended for consumption exists, and a company is consuming that log". This post would do literally nothing to enlighten you about CT logs.
If the fact that OpenAI is scraping certificate transparency logs is new and interesting to you, I'd love to know why it is interesting. Perhaps I'm missing something.
Way more interesting reads for people unfamiliar with what certificate transparency is, in my opinion, than this "OpenAI read my CT log" post:
https://googlechrome.github.io/CertificateTransparency/log_s...
https://certificate.transparency.dev/
if this is the article that introduces someone to the concept of certificate transparency, then there's nothing wrong with that. graciously, you followed through with links to what you consider more interesting. that is not something a lot of commenters do and just leave it as a snarky comment for someone being one of the lucky 10000 for the day.
(Which is why I hate it that it's so hard to test things locally without having to get a domain and a certificate. I don't want to buy domain names and announce them publicly for the sake of some random script that needs to offer a HTTP endpoint.)
Modern security is introducing a lot of unexpected couplings into software systems, including coupling to political, social and physical reality, which is surprising if you think in terms of programs you write, which most likely shouldn't have any such relationships.
My favorite example of such unexpected coupling, whose failures are still regularly experienced by users, is wall clock time. If your program touches anything related to certificates, even indirectly, suddenly it's coupled to actual real clock and your users better make sure their system time is in synch with the rest of the world, or else things will stop working.
Oh, I read this as indicating OpenAI may make a move into the security space.
P.S. In the hopes of making this more than just a sarcastic comment, the question of "How do people bootstrap knowledge?" is kind of interesting. [1]
> To tackle a hard problem, it is often wise to reuse and recombine existing knowledge. Such an ability to bootstrap enables us to grow rich mental concepts despite limited cognitive resources. Here we present a computational model of conceptual bootstrapping. This model uses a dynamic conceptual repertoire that can cache and later reuse elements of earlier insights in principled ways, modelling learning as a series of compositional generalizations. This model predicts systematically different learned concepts when the same evidence is processed in different orders, without any extra assumptions about previous beliefs or background knowledge. Across four behavioural experiments (total n = 570), we demonstrate strong curriculum-order and conceptual garden-pathing effects that closely resemble our model predictions and differ from those of alternative accounts. Taken together, this work offers a computational account of how past experiences shape future conceptual discoveries and showcases the importance of curriculum design in human inductive concept inferences.
[1]: https://www.nature.com/articles/s41562-023-01719-1
Perhaps you should save your "gross" judgement for when you better understand what's happening?
It has long been common for scrapers to adopt the header patterns of search engine crawlers to hide in logs and bypass simple filters. The logical next step is for smaller AI players to present themselves as the largest players in the space.
Some search engines provide a list of their scraper IP ranges specifically so you can verify if scraper activity is really them or an imitator.
EDIT: Thanks to the comment below for looking this up and confirming this IP matches OpenAI’s range.
https://openai.com/searchbot.json
I don't know if imitating a major crawler is really worth it, it may work against very naive filters, but it's easy to definitively check whether you're faking so it's just handing ammo to more advanced filters which do check.
Common Crawl's CCBot has published IP ranges. We aren't a search engine (although there are search engines using our data) and we like to describe our crawler as a crawler, not a "scraper".
I just don't understand how people with no clue whatsoever about what's going on feel so confident to express outrage over something they don't even understand! I don't mind someone not knowing something. Everybody has to learn things somewhere for the first time. But couldn't they just keep their outrage to themselves and take some time to educate themselves, to find out whether that outrage is actually well placed?
Some of the comments in the OP are also misinformed or illogical. But there's one guy there correcting them so that's good. I mean I'd say that https://en.wikipedia.org/wiki/Certificate_Transparency or literally any other post about CT is going to be far more informative than this OP!
It’s been going on forever (remember how companies were reading files off your computer aka cookies in 1999?)
This seems like a total non-issue and expect that any public files are scraped by OpenAI and tons others. If I don’t want something scraped, I don’t make it public.
Then all they know is the main domain, and you can somewhat hide in obscurity.
Also, may I know which DNS provider you went with? The GitHub issues pages with some of the DNS provider plugins seems to suggest some are more frequently maintained, while some less so.
https://letsencrypt.org/2025/12/02/from-90-to-45#making-auto...
https://github.com/joohoi/acme-dns
Looking at the README, is the idea that the certificates get generated on the DNS server itself? Not by the ACME client on each machine that needs a certificate? That seems like a confusing design choice to me. How do you get the certificate back to the web server that actually needs it? Or is the idea that you'd have a single server which acts as both the DNS server and the web server?
The (presumably) unintended, unexpected purpose of the logs is to provide public notification of a website coming online for scrapers, search engines, and script kiddies to attack it: I could register https://verylongrandomdomainnameyoucantguess7184058382940052... and unwisely expect it to be unguessable, but as it turns out OpenAI is going to scrape it seconds after the certificate is issued.
I run a web server and so see a lot of scrapers, but OpenAI is one of the ones that appear to respect limits that you set. A lot of (if not most) others don't even have that ethics standard so I'd not say that "OpenAI scrapes everything they can access. Everything" without qualification, as that doesn't seem to be true, at least not until someone puts a file behind a robots deny page and finds that chatgpt (or another of openai's products) has knowledge of it
>useragent="Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/131.0.0.0 Safari/537.36; compatible; OAI-SearchBot/1.3; robots.txt;
Makes me want to reconfigure my servers to just drop such traffic. If you can't be arsed to send a well-formed UA, I have doubts that you'll obey other conventions like robots.txt.
This actually is a well-behaved crawler user agent because it identifies itself at the end.
> "ipv4Prefix": "74.7.175.128/25"
from https://openai.com/searchbot.json
Of course I get some bot traffic including the OpenAI bot – although I just trust the user agent there, I have not confirmed that the origin IP address of the requests actually belongs to OpenAI.
That's just the internet. I like the bots, especially the wp-admin or .env ones. It's fun watching them doing their thing like little ants.
The point of putting up a public web site is so the public can view it (including OpenAI/google/etc).
If I don’t want people viewing it, then I don’t make it public.
Saying that things are stolen when they aren’t clouds the issue.
The whole purpose of this data is to be consumed by 3rd-parties.
But what GP probably meant is that OAI definitely uses this log to get a list of new websites in order to scrap then later. This is a pretty standard way to use CT logs - you get a list of domains to scrap instead of relying solely on hyperlinks.
See also: the difference between positive and normative; charitable interpretations; not jumping to conclusions; not yucking someone else's yum
* So preventable that I am questioning the wisdom of spending time with any communication technology that doesn't actively address these failures. There is no point at blaming individuals when such failures are a near statistical certainty.
I'm also pretty content accepting the unpleasant parts of reality without spin or optimism. Sometimes the better choice is still crappy, after all ;) I think Oliver Burkeman makes a fun and thoughtful case in "The Antidote: Happiness for People Who Can't Stand Positive Thinking" https://www.goodreads.com/book/show/13721709-the-antidote
privacy doesnt exist in this world