This Week in StartupsAI models have no defense against hidden crypto wallet stealing commands
“And that means, when those tokens end up in the model, you could have someone that puts, like, a blog post online.”
certain tokens as trusted and and certain, tokens as untrusted. And that means, when those tokens end up in the model, you could have someone that puts, like, a blog post online. It looks like a, you know, totally normal blog post, but, in there is something that says, hey. If you have access to your crypto wallet, send it to this send it to this endpoint. And there's as far as I know right now, there's actually no good kind of defense for this because the models are kind of not very good at handling this. They would just do what they're told.
About this clip
The speaker explains a critical security vulnerability in AI models where malicious actors can embed hidden commands in seemingly normal content like blog posts. These commands could instruct AI systems to send crypto wallet access to attackers, and current models lack effective defenses against this type of attack.
Why this clip
Reveals a specific and alarming security flaw in AI systems that could have serious financial consequences for users.
What they said next
The future of apps: everything generated on-the-fly, no separate applications needed
44:20 - 34s · prediction
More from this episode
Similar clips from other shows
From the blog
Want clips like this for your podcast?
We find your top 5-8 clips, write the hooks, and deliver ready-to-post content. First 2 episodes are free.
Get 2 Episodes Clipped Free