Grok users prompted the AI chatbot to identify victims from redacted Epstein Files images.

[bellingcat@mstdn.social]

Grok users prompted the AI chatbot to identify victims from redacted Epstein Files images. Of 31 requests reviewed, Grok generated images for 27, spanning 18 images and one video. X did not respond to requests for comment on Grok's behaviour by time of publication. https://www.youtube.com/shorts/dhD8OKxZBpk

[suedioh@mastodon.social]

@Bellingcat Is anyone using Grok to remove masks and sunglasses?

[cwicseolfor@zeroes.ca]

@SueDiOh @Bellingcat Yes, which only creates a fake face based on statistical averages and sends people chasing after imaginary people satisfied in the belief “we’ll know them when we see them,” instead of prioritizing the actual truth, and accountability for the real murderers and gestapo wannabes. It’s a perfect tool for ensuring there’s no justice for anyone.

[samiamsam@mastodon.social]

@Bellingcat

ffs

what the actual fuck is going on

[harriettmb@toot.wales]

@Bellingcat They aren’t using Grok to generate the names of the men involved - just using it to generate images of abused and trafficked women. This is why Epstein and the network of powerful men exists - so they can control, traffic, abuse and assault women and children. Same for the ex-husband of Gisele Pelicot and others like him.

[apodoxus@mastodon.online]

@Bellingcat Why is only Grok and X mentioned here?

[yearofthebug@mastodon.social]

@Bellingcat wouldn't this work only if the uncensored pictures were already part of the grok training data?

[fiend_unpleasant@mastodon.social]

@Bellingcat So let me get this straight, the users of the most popular child porn creation and dissemination website are using it's A.I. tools to re-victimize people (many of which were children at the time) who we sex trafficked by a convicted pedophile. Sounds about right for the users of X. Everyone on that website is a pedophile or is ok with pedophilia.

[jorismeys@mstdn.social]

@apodoxus
Because other models have safe guards that don't allow this.
@Bellingcat

[passwordsarehard4@mastodon.social]

@cwicseolfor @SueDiOh @Bellingcat and nobody needs to feel accountable. ICE snatches up an innocent citizen? Well, the algorithm had a bad hit. The mob ruins some poor smucks life? Well, we just ahead of ourselves.

[erikcats@dice.camp]

@Bellingcat oh FFS can't wait until the EU decides pedoworld will no longer be allowed to be accessed from within its borders.

[apodoxus@mastodon.online]

@JorisMeys

Do they really? I find it hard to believe that they aren't trivial to circumvent.

@Bellingcat

[bhm@mastodon.world]

@yearofthebug @Bellingcat Yes, it’s just inpainting random faces. Unless the real images are in the training database or very similar images of the same person(s).

[jorismeys@mstdn.social]

@apodoxus
Not saying they're safe. But eejits that don't understand grok or any other LLM will never be able to "uncensor", are not going to put effort in learning how to circumvent, how trivial it might seem to anyone with basic infosec literacy
@Bellingcat