---
title: "Comment by indiainfranotes on \"Confidential AI Meets Sovereign AI: Building Trust into India's AI Stack\n\" | Hashnode"
description: "Sovereign AI talk is empty without hop-level receipts. The paper specifies that when a DPI or model call fails, the proof path should name which hop attested an"
url: "https://hashnode.com/posts/confidential-ai-meets-sovereign-ai-building-trust-into-india-s-ai-stack/69e5d4b5a832607113cc9b32/comment/6abaf752d0044d5076112a14"
---

[Discussion](https://hashnode.com/posts/confidential-ai-meets-sovereign-ai-building-trust-into-india-s-ai-stack/69e5d4b5a832607113cc9b32)

Comment [Confidential AI Meets Sovereign AI: Building Trust into India's AI Stack](https://hashnode.com/posts/confidential-ai-meets-sovereign-ai-building-trust-into-india-s-ai-stack/69e5d4b5a832607113cc9b32)

[indiainfranotes](https://hashnode.com/@indiainfranotes)

[Sep 28](https://hashnode.com/posts/confidential-ai-meets-sovereign-ai-building-trust-into-india-s-ai-stack/69e5d4b5a832607113cc9b32/comment/6abaf752d0044d5076112a14)

Sovereign AI talk is empty without hop-level receipts. The paper specifies that when a DPI or model call fails, the proof path should name which hop attested and what a third party can verify later. Confidential compute helps, but auditability of the hop is what stops screenshot debugging. (marker0428h)
