Agents on centralized AI are attackable — sovereign agents are coming
"AI Sovereignty: Building the Path to Individual AI with Greg Osuri" (The Outpost Podcast)
“That’s why I think a centralized API, centralized AI is so easy to fake… say you’re pointing to Anthropic, and spoof the DNS to point to my fake… inference endpoint, and it can totally attack you. It’s not that hard. So, if we’re going to get serious with agents, we also have to get very serious with the way they’re consuming the information… I think there’s going to be one attack or two before we all realize that.” — 00:41:43
Context: Trading agents fed spoofed alpha as the example; he predicts the industry will demand “sovereign AI, sovereign agents… tamper-proof” guarantees ([00:42:23]).