Location Verification for AI Compute

The Idea

Prove where AI computation physically occurs. For international agreements (e.g., Gulf negotiations), parties need assurance that compute nominally located in a “neutral” or “verified” jurisdiction is actually running there, not secretly redirected to an unmonitored facility.

Location verification could use: GPS/GNSS with anti-spoofing, network latency triangulation, physical inspection regimes, or cryptographic attestation tied to hardware installed at known locations. The challenge is making this robust against nation-state adversaries who control local infrastructure.

Why It Matters

Many verification schemes assume you know where the hardware is. “Inference-only clusters in third-party jurisdictions” or “training hubs under joint monitoring” only work if location claims are credible. Location verification is a foundational primitive for international compute governance.

Open Questions

• What’s the threat model? Spoofed GPS? Compromised network infrastructure?
• Can latency measurements distinguish datacenters 100km apart?
• How does location verification interact with secure enclaves?
• What physical inspection regime is practical for continuous verification?

Status

Amodo planning demo for policy stakeholders. Tim Fist advising on specifications.

References

• Gulf negotiations context (policy window 2025)
• Related: RFD 017 (Inference-Only Verification Package)