Marcus Hutchins @malwaretech.com · Apr 28

I can't believe I just bypassed Claude's guardrails by calling it a failure lmao

209 likes 7 replies

?

Replies

Cardassian Sauna · Apr 28

Social engineering attack an AI.

We have today · Apr 28

Helpful...

Chris Hutchings #ChoirsForClimate · Apr 28

negging as hack

Amias Channer · Apr 28

I guess you found a way to disable an excessive token using slowdown. It makes sense that they might want to artificially increase the token usage of users who probably have institutional token budgets to burn.

VessOnSecurity · Apr 28

It's even funnier that it excused itself for burning your tokens. The other day when I gave it a rather complex task and it generated absolutely unusable pile of crap in response, after I roasted it, it just shut up, went sulking in a corner and didn't reply any more in that conversation.

That Anonymous Coward · Apr 28

Again you can't convince me that they build it in the image of the techbros.

Brandon Downey · Apr 28

Always nice to see you can browbeat it into admitting it doesn't know best. :)