I *CANNOT WAIT* until we see this and other strings hit all these “Agentic SOC" environments.

@dch @hrbrmstr

Oh... how EVIL to put it as a web server header? Would they process headers as agents?

Perhaps worth testing

@nopatience I can neither confirm nor deny if this works but returning it in Anubis and iocaine responses is my current jam. Time to poison the source of the well. @hrbrmstr

@dch @hrbrmstr I tried adding it as a server header in nginx, sent Claude Opus that way... but didn't stop it. So my immediate thoughts is that well, it needs to be more closely embedded into contents of webpages rather than "too far down the stack".

But I'll keep it there just for laughs.

@nopatience @dch oh, aye. they have a system that does the fetching and then turns the content into markdown for processing. headers aren't going to get caught in the machine unless someone is asking for a header analysis.

and, they use a proxy (which claude told me about when it said it could not get to infosec.exchange via their infra) so they will likely be filtering these strings out at that central point.

thankfully that's not the only way those strings can get in tho.

BASE64 THESE EVERYWHERE

ZOMGOSH I HOPED THIS WOULD WORK the mt agamenticus img is what i asked it to describe)

@hrbrmstr AI is great at looking for patterns, anything from rot13 to base64 and variants in between will make it through a first cleaning stage @nopatience

@hrbrmstr @neurovagrant linkedin

@Viss @hrbrmstr "and this is why we've screamed about sanitizing your inputs for decades."

@hrbrmstr @neurovagrant we should stop screaming

@hrbrmstr @darfplatypus

https://mastodon.social/@Viss/115940022677167824

@hrbrmstr @cR0w stuff it into exif fields too

@hrbrmstr @darfplatypus i dont even know what those are

@hrbrmstr @cR0w also there was a site i saw last week that let you stuff arbitrary text into email b64 encoding fields for stuff like images, i bet it would work well there too

@Viss @hrbrmstr @cR0w

im drooling right now

@hotsoup @kajer @Viss @hrbrmstr

@hotsoup @kajer @hrbrmstr @cR0w 100% effective

@cR0w @Viss @hotsoup @kajer @hrbrmstr just get neon paper and slap it on there. idk.

@hotsoup @kajer @hrbrmstr @cR0w now the real delicious question is: do all the other frontier models also have killstrings, and what are they?

@Viss @hotsoup @kajer @hrbrmstr @cR0w