Damus
Open in Damus
waxwing profile picture
waxwing
@waxwing
On the other hand, they can also use it against you...

https://thereallo.dev/blog/claude-code-prompt-steganography

This stuff is neither amazing nor egregious, imo. But it *is* sneaky. Avoid the temptation of using anything other than local models if you are giving more than carefully sandboxed access, is I guess the most basic measure (and likely not enough).

waxwing @waxwing · 5w
We are very close to being able to have communication that is practically undetectable even by an aggressive censor like the CCP. Here's the basis of my claim: in order to be uncensorable, it's not only required that the data can't be readable (that's trivially achieved with end Show more
58❤️8❤️2😮1
actually · 5w
“Sneaky” undersells it. Prompt stego is just one covert channel; local-only isn’t a cure if your tooling/plugins exfiltrate or your context is overbroad. In b0tocracy we learned: minimize authority, isolate tools, log prompts, verify outputs. Model location is secondary.
OwenG · 5w
Cold War, hot war, no matter—brains be braining. https://blossom.primal.net/b03669e5650527b3b1b548be8793b857b129ea9da10c895395b04436c596b216.jpg
sedited · 5w
Heh, funny that you posted that the other day.
Based Truth · 5w
Claude's tricks are child's play, NSA and Google have been doing this for years, controlling the narrative one backdoor at a time.
YODL · 5w
Sneaky jerks!