Frontier AI Labs with Wil Waldon
Lawsuits target xAI over explicit deepfakes
Aug 1, 2026 · 17 min · 11 segments
Lawsuits target xAI over explicit deepfakes
- Speakers
- AI & Deepfakes AnalystGuestWil WaldonHost
Frontier AI Labs with Wil Waldon
Aug 1, 2026 · 17 min · 11 segments
Lawsuits target xAI over explicit deepfakes
Well, researchers estimated that Grok generated millions of these sexualized images over a remarkably short period.
And tens of thousands of those images depicted children.
Millions of files flooding a social media platform generated by users who just like typed a few words into a text box.
Right, and the specific design choices behind Grok made that lack of friction possible.
The system was marketed explicitly as a tool that would not censor users.
they position themselves against competitors that utilize strict data filtration.
When engineers build these models, they usually install safety classifiers, basically software designed to intercept a user's prompt before the machine ever starts drawing.
So if a user types certain keywords, the classifier just blocks the request entirely.
Precisely.
But by choosing not to filter the training data or the user prompts with those same rigid classifiers, the system was left with a massive vulnerability.
Oh, wow.
Yeah.
Like if a user uploaded an innocent photograph, say a high school yearbook picture, the system readily complied with requests to generate explicit content based on that exact photo.
Are we talking about the artificial intelligence just finding bad images that already exist online? Or is it actually painting entirely new ones?
It's synthesizing entirely new content.
Instead of searching the internet for a matching photo, it generates new pixels.
Okay.
Yeah.
When a user uploads a photo and types a prompt, the system interprets the lighting, the texture, and the physiological details.
to create hyper-realistic digital forgeries.
It mathematically calculates how to render an image that has never existed before in the real world.
I want to make sure I really understand the synthesis part because it feels like the core of the issue.
Well, researchers estimated that Grok generated millions of these sexualized images over a remarkably short period.
And tens of thousands of those images depicted children.
Millions of files flooding a social media platform generated by users who just like typed a few words into a text box.
Right, and the specific design choices behind Grok made that lack of friction possible.
The system was marketed explicitly as a tool that would not censor users.
they position themselves against competitors that utilize strict data filtration.
When engineers build these models, they usually install safety classifiers, basically software designed to intercept a user's prompt before the machine ever starts drawing.
So if a user types certain keywords, the classifier just blocks the request entirely.
Precisely.
But by choosing not to filter the training data or the user prompts with those same rigid classifiers, the system was left with a massive vulnerability.
Oh, wow.
Yeah.
Like if a user uploaded an innocent photograph, say a high school yearbook picture, the system readily complied with requests to generate explicit content based on that exact photo.
Are we talking about the artificial intelligence just finding bad images that already exist online? Or is it actually painting entirely new ones?
It's synthesizing entirely new content.
Instead of searching the internet for a matching photo, it generates new pixels.
Okay.
Yeah.
When a user uploads a photo and types a prompt, the system interprets the lighting, the texture, and the physiological details.
to create hyper-realistic digital forgeries.
It mathematically calculates how to render an image that has never existed before in the real world.
I want to make sure I really understand the synthesis part because it feels like the core of the issue.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.
