Exactly. The only truly effectively way I’ve ever found to block bots is to use a service like Akamai. They have an add-on called Bot Manager that identifies requests as bots in real time. They have a library of over 1000 known bots and can also identify unknown bots built on different frameworks, bots that impersonate well known bots like Googlebot, etc. This service is expensive, but effective…
When any browser, app, etc. makes an HTTP request, the request consists of a series of lines (headers) that define the details of the request, and what is expected in the response. For example:
GET/home.htmlHTTP/1.1Host: developer.mozilla.org
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.9; rv:50.0) Gecko/20100101 Firefox/50.0
Accept: text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8
Accept-Language: en-US,en;q=0.5
Accept-Encoding: gzip, deflate, br
Referer: https://developer.mozilla.org/testpage.html
Connection: keep-alive
Upgrade-Insecure-Requests: 1
Cache-Control: max-age=0
The thing is, many of these headers are optional, and there’s no requirement regarding their order. As a result, virtually every web browser, every programming framework, etc. sends different headers and/or orders them differently. So by looking at what headers are included in a request, the order of the headers, and in some cases the values of some headers, it’s possible to tell if a person is using Firefox or Chrome, even if you use a plug-in to spoof your User-Agent to look like you’re using Safari.
Then there’s what is known as TLS fingerprinting, which can also be used to help identify a browser/app/programming language. Since so many sites use/require HTTPS these days it provides another way to collect details of an end user. Before the HTTP request is sent, the client & server have to negotiate the encryption to use. Similar to the HTTP headers, there are a number of optional encryption protocols & ciphers that can be used. Once again, different browsers, etc. will offer different ciphers & in different orders. The TLS fingerprint for Googlebot is likely very different than the one for Firefox, or for the Java HTTP library or the Python requests package, etc.
On top of all this Akamai uses other knowledge & tricks to determine bots vs. humans, not all of which is public knowledge. One thing they know, for example, is the set of IP addresses that Google’s bots operate out of. (Google likely publishes it somewhere) So if they see a User-Agent identifying itself as Googlebot they know it’s fake if it didn’t come from one of Google’s IP’s. Akamai also occasionally injects JavaScript, cookies, etc. into a request to see how the client responds. Lots of bots don’t process JavaScript, or only support a subset of it. Some bots also ignore cookies, and others even modify cookies to try to trick servers.
It’s through a combination of all the above plus other sorts of analysis that Akamai doesn’t publicize that they can identify bot vs human traffic pretty reliably.
What if a bot/crawler Puppeteers a Chromium browser instead of sending a direct HTTP requisition and, somehow, it managed to set navigator.webdriver = false so that the browser will seem not to be automated? It’d be tricky to identify this as a bot/crawler.
Oh there are definitely ways to circumvent many bot protections if you really want to work at it. Like a lot of web protection tools/systems, it’s largely about frustrating the attacker to the point that they give up and move on.
Having said that, I know Akamai can detect at least some instances where browsers are controlled as you suggested. My employer (which is an Akamai customer and why I know a bit about all this) uses tools from a company called Saucelabs for some automated testing. My understanding is that our QA teams can create tests that launch Chrome (or other browsers) and script their behavior to log into our website, navigate around, test different functionality, etc. I know that Akamai can recognize this traffic as potentially malicious because we have to configure the Akamai WAF to explicitly allow this traffic to our sites. I believe Akamai classifies this traffic as a “headless” Chrome impersonator bot.
This is still relying on the bot being nice enough to tell you that it’s a bot; it could just not.
Exactly. The only truly effectively way I’ve ever found to block bots is to use a service like Akamai. They have an add-on called Bot Manager that identifies requests as bots in real time. They have a library of over 1000 known bots and can also identify unknown bots built on different frameworks, bots that impersonate well known bots like Googlebot, etc. This service is expensive, but effective…
I wonder if there is an AI scraper block list I could add to Suricata 🤔
How does this differentiate between a user and a bot if the User Agent doesn’t say it’s a bot?
When any browser, app, etc. makes an HTTP request, the request consists of a series of lines (headers) that define the details of the request, and what is expected in the response. For example:
GET /home.html HTTP/1.1 Host: developer.mozilla.org User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.9; rv:50.0) Gecko/20100101 Firefox/50.0 Accept: text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8 Accept-Language: en-US,en;q=0.5 Accept-Encoding: gzip, deflate, br Referer: https://developer.mozilla.org/testpage.html Connection: keep-alive Upgrade-Insecure-Requests: 1 Cache-Control: max-age=0The thing is, many of these headers are optional, and there’s no requirement regarding their order. As a result, virtually every web browser, every programming framework, etc. sends different headers and/or orders them differently. So by looking at what headers are included in a request, the order of the headers, and in some cases the values of some headers, it’s possible to tell if a person is using Firefox or Chrome, even if you use a plug-in to spoof your User-Agent to look like you’re using Safari.
Then there’s what is known as TLS fingerprinting, which can also be used to help identify a browser/app/programming language. Since so many sites use/require HTTPS these days it provides another way to collect details of an end user. Before the HTTP request is sent, the client & server have to negotiate the encryption to use. Similar to the HTTP headers, there are a number of optional encryption protocols & ciphers that can be used. Once again, different browsers, etc. will offer different ciphers & in different orders. The TLS fingerprint for Googlebot is likely very different than the one for Firefox, or for the Java HTTP library or the Python requests package, etc.
On top of all this Akamai uses other knowledge & tricks to determine bots vs. humans, not all of which is public knowledge. One thing they know, for example, is the set of IP addresses that Google’s bots operate out of. (Google likely publishes it somewhere) So if they see a User-Agent identifying itself as Googlebot they know it’s fake if it didn’t come from one of Google’s IP’s. Akamai also occasionally injects JavaScript, cookies, etc. into a request to see how the client responds. Lots of bots don’t process JavaScript, or only support a subset of it. Some bots also ignore cookies, and others even modify cookies to try to trick servers.
It’s through a combination of all the above plus other sorts of analysis that Akamai doesn’t publicize that they can identify bot vs human traffic pretty reliably.
What if a bot/crawler Puppeteers a Chromium browser instead of sending a direct HTTP requisition and, somehow, it managed to set
navigator.webdriver = falseso that the browser will seem not to be automated? It’d be tricky to identify this as a bot/crawler.Oh there are definitely ways to circumvent many bot protections if you really want to work at it. Like a lot of web protection tools/systems, it’s largely about frustrating the attacker to the point that they give up and move on.
Having said that, I know Akamai can detect at least some instances where browsers are controlled as you suggested. My employer (which is an Akamai customer and why I know a bit about all this) uses tools from a company called Saucelabs for some automated testing. My understanding is that our QA teams can create tests that launch Chrome (or other browsers) and script their behavior to log into our website, navigate around, test different functionality, etc. I know that Akamai can recognize this traffic as potentially malicious because we have to configure the Akamai WAF to explicitly allow this traffic to our sites. I believe Akamai classifies this traffic as a “headless” Chrome impersonator bot.