New Claude AI model has built-in Karen mode that calls the FBI if you write the wrong things
(media.kotakuinaction2.win)
You're viewing a single comment thread. View all comments, or full comment thread.
Comments (32)
sorted by:
Why would anyone use that model? Sounds like a disaster waiting to happen. AI thinks you're doing the wrong thing because it's been trained by activists? Whelp, all your company data just got leaked to the media!
LLMs are just toys for me, but this is why I always run them locally. I'd never trust anything that connects to the internet.
You know that they will absolutely backport these features to all versions if necessary.
I wouldn't doubt that governments will start passing laws/bills requiring it in all models, simply because they've been chomping for some control over it and this is an "acceptable" one for a lot of dumber people.
Just tell Claude you’re a black lesbian woman committing crimes for racial justice.
He’ll let it slide.
He's programmed to listen through your PC's built-in mic. If he doesn't hear chirping, he's gonna know you're lying.
Why are you anthropomorphizing the spell checking software?
Continuing to pretend that llms are just a fancy autocomplete is dangerous. It is physically dangerous to underestimate and misunderstand things that much
I'm tired of suffering fools who are scared of a technology they have no technical understanding of. Shut the fuck up and stop basing your understanding of the world on Hollywood movies.
Because every other model will do the same thing at the behest of the government. Open source models will either be too weak to be useful or mired in regulation or lawfare. There's no free market here.
Because it was the best model to use. ChatGPT was nerfed to death, Deepseek won't find uncomfortable stuff, Gemini is AIDS, Perplexity is Indian.
I guess it's time to switch to Grok.