Are hosting providers seeing more AI-assisted security testing?

RackWorks

New member
Curious what others in the hosting industry are seeing lately. Are you noticing an increase in AI-assisted vulnerability scanning, automated pentesting, or scanners that go beyond generic probes and actually test application logic?

Wondering whether this is becoming more common across hosting environments, or if it's just business as usual from a security perspective.
 
Honestly, from raw access logs alone it's pretty tough to tell an LLM-driven scanner apart from a tuned Nuclei script or headless browser bot.

The real tell for me is stateful behavior.. like whether an IP actually holds onto session cookies, parses responses, and chains multi-step requests together instead of just spraying the same .env and wp-login paths across hundreds of URLs.

Volume spikes are mostly business as usual at this point, so raw request counts don't mean much on their own. Are you guys seeing more stateful probing on your own nodes lately, or still mostly the usual background noise?
 
What hit us didn’t act like a typical scanner. Block a scanner’s IP and it’s usually gone. This came right back from a new IP with a different user agent, picked up where it left off and went after the same parts of the application. Every block we put in place, it worked around. It turned into a game of whack-a-mole.

IP and user agent rotation on its own is nothing new. Botnets and proxy pools have done it for years. What I hadn’t seen before from conventional automation was something that reacted to the responses it got and changed its approach in real time.

Can I prove from an access log that an LLM was behind it? No, but nobody can prove it wasn’t, either. Based on how it behaved, I’m comfortable calling it AI-driven. Curious whether anyone else has run into the same pattern.
 
What hit us didn’t act like a typical scanner. Block a scanner’s IP and it’s usually gone. This came right back from a new IP with a different user agent, picked up where it left off and went after the same parts of the application. Every block we put in place, it worked around. It turned into a game of whack-a-mole.

IP and user agent rotation on its own is nothing new. Botnets and proxy pools have done it for years. What I hadn’t seen before from conventional automation was something that reacted to the responses it got and changed its approach in real time.

Can I prove from an access log that an LLM was behind it? No, but nobody can prove it wasn’t, either. Based on how it behaved, I’m comfortable calling it AI-driven. Curious whether anyone else has run into the same pattern.
Picking right back up from the exact step where the last IP got blocked is wild.. definitely means there's a central orchestrator keeping state behind that proxy pool imo.

At that point IP blocks really are pure whack-a-mole. Checking whether those rotated IPs still share the same TLS fingerprint (like JA3/JA4) or header order sometimes gives you something steadier to hook a WAF rule onto, otherwise rate-limiting the actual app workflow itself is about the only thing that slows them down.
 
Back
Top