Every server with port 80/443 open has thousands of hits a day from random boxes looking for wordpress login pages. The only new thing is that they're pretending to be a different type of annoying bot. There's a new layer of sophistication and subterfuge, but it's the same junk traffic we've always dealt with.
Think about how many webmaster and business owners' egos are stroked by all the traffic they are getting, when in actuality they are often just serving thousands of bots.
People or their agents must be accidentally committing or publishing their repository level secrets and configs with enough regularity that it’s worth scanning.
There are a few novel ones but I’ve been seeing most of them in my logs for longer than generative AI has existed. This isn’t remotely new, the vector is just getting bigger.
On average about 100 (TCP) requests hit my home router per minute doing various probing and scanning. Lots of checking for the telnet port obviously. Sometimes you can see a swarm of entirely different IPs scanning the full port range (probing the ports one-by-one).
You'll see a lot of deepfield, censys-scanner, visionheight.com, shadowserver.io, etc., but also the usual suspects of Chinese or Russian IPs.
With OpenWRT I use something like this: `tcpdump -i pppoe-wan 'inbound and tcp[tcpflags] & (tcp-syn|tcp-ack) == tcp-syn'`, or alternatively `tcpdump -i pppoe-wan 'inbound and tcp[tcpflags] & (tcp-syn|tcp-ack) == tcp-syn and not port 44000'`, if we have some torrent client running (e.g. here at port 44000) which would mess up the result. I'm not sure it's the best way to handle this, but it's definitely enlightening what bounces off on the router.
I have ubiquity UniFi for this reason (amongst others). OpenWRT is a good choice as well. Most home router software is such junk, might as well leave the door open..
The easiest way to deal with the usual suspects is to just block the entire countries network range(s). There really is no reason they should be connecting to your home router anyway, and you lose nothing from blocking them.
Sure their packets will still hit your router, but if they are dropped immediately at least you're not wasting a syn-ack on them.
Many of those user-agents listed are often faked. Look up which ASN owns their IP. If I block most VPS providers most of the faked bots vanish. There are still some running from residential and phones using hijacked code (readers that are not really just readers but really multipurpose proxies). On that note, do not trust the linked source code but rather decompile the live code your phone is running and have AI analyze it.
Is there an easy way to block any requests originating from VPS etc instead of residential/commercial IP from legitimate users ? I know cloudflare does a few things but I really want to figure out a way to block any request say at nginx or caddy (reverse proxy) from reaching origin servers if they are not from an IP that is not a VPS etc.
Yes but it’s not cheap. Maxmind and ipinfo etc sell a tier that tells you this information, then you can 403 based on it.
But the price is nuts like $40,000 a year.
Yeah, that's exactly what these visits are: faked user agents that fail IP verification or Web Bot Auth. What's interesting is the surge across so many websites in the last week.
There are many possibilities but one of them could be some new vuln was released and they are looking for it. That would require looking at the URL's they are requesting. Botters run their own purpose built campaigns. Do you also have a summary of URL's requested by unique counts?
Same for the origin IP address. The fiber leaving your country is tapped, and those people can inject packets with any origin IP that they want. Your ISP has no way to check if their peer actually received a certain packet from a certain country or not.
From a technical perspective, all this "china/russia" attribution is built on a quite shaky foundation. As a sysadmin you'd never know if it would be the British crown attacking your European company instead.
Not minimizing nation state cyber crime here, but the packet goes through many hands with different incentives.
Problem here is there are not single fibers attaching (most) countries, but a bunch of them. If you control both the ingress and egress for some particular users it's possible, but if you don't then your probing packing may end up back in China with a lot of evidence of backscatter.
This kind of stuff is getting pretty wild, even for using something like Cloudflare it seems like a good idea to have another layer behind it that's non-cloudflare for when vulnerabilities are discovered.
Definitely, the point here though is there is a stat sig surge in the chart in the last week, across thousands of websites. At least a surge in this particular spoofing pattern.
This is a random sample of completely unrelated websites, which indicates that the total scale is much larger. This is not saying that it is difficult to make thousands of requests.
Is this your company? If it is, your cheerleading makes you very hard to trust. If it’s not, I’m sure that everyone gets the point - you adore everything about this research and can’t see any possible problems.
I think this is more of a "if you left your AI tools exposed someone is looking for them" change. Hacking someone else's agents sounds like a great way to spend less on your own tokens.
Every server with port 80/443 open has thousands of hits a day from random boxes looking for wordpress login pages. The only new thing is that they're pretending to be a different type of annoying bot. There's a new layer of sophistication and subterfuge, but it's the same junk traffic we've always dealt with.
Think about how many webmaster and business owners' egos are stroked by all the traffic they are getting, when in actuality they are often just serving thousands of bots.
Opening port 80 and realizing the world is an anarchic warzone is a canonic rite of initiation for otherwise innocent backend devs and sysadmins.
Another interesting thing here is the paths they're targeting, many are for newish AI coding tools
People or their agents must be accidentally committing or publishing their repository level secrets and configs with enough regularity that it’s worth scanning.
Totally. I'm sure this campaign was inspired by sloppy vibe coding
There are a few novel ones but I’ve been seeing most of them in my logs for longer than generative AI has existed. This isn’t remotely new, the vector is just getting bigger.
On average about 100 (TCP) requests hit my home router per minute doing various probing and scanning. Lots of checking for the telnet port obviously. Sometimes you can see a swarm of entirely different IPs scanning the full port range (probing the ports one-by-one).
You'll see a lot of deepfield, censys-scanner, visionheight.com, shadowserver.io, etc., but also the usual suspects of Chinese or Russian IPs.
With OpenWRT I use something like this: `tcpdump -i pppoe-wan 'inbound and tcp[tcpflags] & (tcp-syn|tcp-ack) == tcp-syn'`, or alternatively `tcpdump -i pppoe-wan 'inbound and tcp[tcpflags] & (tcp-syn|tcp-ack) == tcp-syn and not port 44000'`, if we have some torrent client running (e.g. here at port 44000) which would mess up the result. I'm not sure it's the best way to handle this, but it's definitely enlightening what bounces off on the router.
I have ubiquity UniFi for this reason (amongst others). OpenWRT is a good choice as well. Most home router software is such junk, might as well leave the door open..
The easiest way to deal with the usual suspects is to just block the entire countries network range(s). There really is no reason they should be connecting to your home router anyway, and you lose nothing from blocking them.
Sure their packets will still hit your router, but if they are dropped immediately at least you're not wasting a syn-ack on them.
I can corroborate visionheight and shadowserver from my firewall logs.
I recently blogged about some Cloudflare Workers I developed to combat this type of traffic: https://code.backwater.systems/blog/#2026-06-29T23:40:00.000...
Many of those user-agents listed are often faked. Look up which ASN owns their IP. If I block most VPS providers most of the faked bots vanish. There are still some running from residential and phones using hijacked code (readers that are not really just readers but really multipurpose proxies). On that note, do not trust the linked source code but rather decompile the live code your phone is running and have AI analyze it.
Is there an easy way to block any requests originating from VPS etc instead of residential/commercial IP from legitimate users ? I know cloudflare does a few things but I really want to figure out a way to block any request say at nginx or caddy (reverse proxy) from reaching origin servers if they are not from an IP that is not a VPS etc.
I run an open source honeypot that collects these botnet scans and produces blocklists.
Blocklist download and configuration: https://knock-knock.net/blocklist
Honeypot dashboard, where you can see attempted attacks in realtime: http://knock-knock.net
API: http://knock-knock.net/api
> /commercial IP from legitimate users
No, because legitimate users do not just use residential and "commercial" IPs. Like me, right now
You are the 0.001%
Much more than 0.001% of people care about their privacy or (the larger portion) do not have unfiltered access to the internet.
Yes but it’s not cheap. Maxmind and ipinfo etc sell a tier that tells you this information, then you can 403 based on it. But the price is nuts like $40,000 a year.
You don't need to spend anywhere near $40k a year to get that info... You don't even need to spend $1
Yeah, that's exactly what these visits are: faked user agents that fail IP verification or Web Bot Auth. What's interesting is the surge across so many websites in the last week.
There are many possibilities but one of them could be some new vuln was released and they are looking for it. That would require looking at the URL's they are requesting. Botters run their own purpose built campaigns. Do you also have a summary of URL's requested by unique counts?
Looks like many of the paths relate to AI coding tools. There are some examples below the chart
You keep repeating this about a small minority of the tools that were posted.
Same for the origin IP address. The fiber leaving your country is tapped, and those people can inject packets with any origin IP that they want. Your ISP has no way to check if their peer actually received a certain packet from a certain country or not.
From a technical perspective, all this "china/russia" attribution is built on a quite shaky foundation. As a sysadmin you'd never know if it would be the British crown attacking your European company instead.
Not minimizing nation state cyber crime here, but the packet goes through many hands with different incentives.
Problem here is there are not single fibers attaching (most) countries, but a bunch of them. If you control both the ingress and egress for some particular users it's possible, but if you don't then your probing packing may end up back in China with a lot of evidence of backscatter.
This kind of stuff is getting pretty wild, even for using something like Cloudflare it seems like a good idea to have another layer behind it that's non-cloudflare for when vulnerabilities are discovered.
Someone is always running mass vulnerability scans. That's a "water is wet" state of the Internet.
Definitely, the point here though is there is a stat sig surge in the chart in the last week, across thousands of websites. At least a surge in this particular spoofing pattern.
It's still not really anything special. Thousands isn't even large scale.
Any random bozo can trigger that.
This is a random sample of completely unrelated websites, which indicates that the total scale is much larger. This is not saying that it is difficult to make thousands of requests.
Is this your company? If it is, your cheerleading makes you very hard to trust. If it’s not, I’m sure that everyone gets the point - you adore everything about this research and can’t see any possible problems.
Not asking for trust, just sharing the data/math
I think this is more of a "if you left your AI tools exposed someone is looking for them" change. Hacking someone else's agents sounds like a great way to spend less on your own tokens.
Nah these are just the typical vuln scans that try all kinda basic shit, looking for anything commonly used AI tools included.
Such a great idea that a sufficiently advanced LLM might take it upon itself to do so, and hide the truth from you.