Le 28/11/2025 à 05:02, onefang a écrit :
> On 2025-11-27 12:05:28, g4sra via Dng wrote:
>> Welcome to the AI future.
>> anthropic.com have been having bad press since last year and don't appear to care how much upset they cause.
> Just as an example, got 404s on
> /pub/mirrors/fedora-buffet/epel/10.1z/Everything/aarch64/Packages/r/rust-mio0.8-devel-0.8.11-2.el10_1.noarch.rpm
> 4727 times in the last 24 hours! Hmmm, AIs have become Rusty? Muahahaha!
>
> Then it pokes around in similar paths for the same Everything, not always
> in /pub/mirrors. Then looks for other things multiple times in my non
> existant Fedora mirror.
>
> This has become Artificial Insanity now. "The definition of insanity is
> doing the same thing over and over again and expecting different
> results." It doesn't matter how many thousands of times per day
> ClaudeBot asks me, I'm not gonna suddenly sprout a Fedora mirror.
>
> What would an LLM even do with an RPM package? "Much ado about nothing."
>
>> Using Agent Headers or robots.txt is pointless as the aggressor cannot be trusted to be honest. I am currently looking for some form of proactive\reactive dynamic firewall blocking for this type of 'attack'.
>> I am also investigating if I can tarpit (without causing myself significant overhead) to actually slow down probes onto my network by causing heightened resource usage at the aggressor's end.
>>
>> If you know of anything that may help please give me a head-up.
> One of my ideas is to delay the 404 report. Normal people probably wont
> even notice if the 404 report took a few seconds. It may slow down the
> bots, but maybe not.
>
> It's an on going arms race. I hope this bubble bursts soon.

    Just an idea: if you could, via the netfilter, route the packets
from this address to another port, where a dedicated server would just
respond to any request with an http redirect to some site you hate 😎
I'd be curious to know if that's practical.

--     Didier