Working paper. Content reflects ongoing research and may change.

← Back to Working Papers
verification offline cryptography draft

Offline Inference Verification: A Hash-Chain Approach

November 12, 2025

Draft note on a hash-chain approach for verifying inference runs in air-gapped environments. We describe a proposed artifact format and signing protocol, with early prototype measurements in progress.

Status: Draft. Prototype measurements are preliminary.

Abstract

Offline AI deployments in high-assurance environments require verification mechanisms that function without network connectivity. Traditional approaches rely on remote attestation or trusted third parties, both incompatible with air-gapped networks.

We outline a hash-chain verification protocol that:

  1. Links each inference run to its predecessor cryptographically
  2. Includes hashes of model weights, adapter state, configuration, and inputs
  3. Signs the resulting proof pack with Ed25519
  4. Enables later verification without the signing key

Methodology

Our approach builds on Merkle tree structures adapted for sequential inference runs. Each proof pack contains:

  • SHA-256 hash of input data
  • SHA-256 hash of model weights at inference time
  • SHA-256 hash of active adapter weights
  • SHA-256 hash of runtime configuration
  • SHA-256 hash of output data
  • Reference to previous proof pack hash
  • Ed25519 signature over the combined structure

Results

Prototype overhead measurements on Apple M2 Pro (internal):

WorkloadBaseline (ms)With Proof (ms)Overhead
Small (1K tokens)1451482.1%
Medium (4K tokens)5805901.7%
Large (16K tokens)234023751.5%

Prototype hash computation and signing add ~3ms overhead in our setup; expect variance.

Conclusion

Hash-chain verification appears practical for offline deployments, with overhead still being characterized. The approach provides tamper evidence without requiring network connectivity.

About MLNavigator Research Group

We explore verifiable, offline AI systems and publish working notes as the research develops. If you want to discuss collaboration or pilots, reach out.

Contact Research Pillars