Interesting. If I understand the claim correctly: If AI agents are going to roam the internet on your behalf, anything they come across could be a prompt injection attack vector. That's an enormous attack surface.
colinlevine•2h ago
Yeah, it's a huge issue. I'm actually building a startup to protect agents from these kinds of attacks. https://savira.dev
AnimalMuppet•2h ago
colinlevine•2h ago