Skip to content
@ai-verification-observatory

ai-verification-observatory

AI Verification Observatory

An effort to foster more open source collaboration in the AI verification ecosystem. We all share the same mission, let's join forces to make sure we accelerate the research and adoption of the verification primitives.

  • Open measurement of systems that prove what an AI model computed: what each proof establishes, under which assumptions, at what cost, and whether anyone other than its authors got the same result.
  • Research experiments
  • Prototype developments
  • Libraries, shared primitives and tooling

Results Benchmark Methodology by RFC Apache-2.0 and CC BY 4.0

Work

Track Scope Status
Inference Proofs Benchmark for provable LLM inference, in the spirit of Ethproofs, with divisions and peer review borrowed from MLPerf. Release candidate v0.1 public at inferenceproofs.ai: SASH zk-inference-prover (Llama 3.1 8B) and FlashProve (Llama 4 Maverick) on their native workloads, with vLLM production-serving reference rows, on NVIDIA H200 141 GB. Every number is at tier reported. Neither claim card has been reviewed by its system's team yet. 16 RFCs, all Draft.
Core library Shared verification components, once benchmarks show which interfaces converge. Not started. Shared code is limited to the harness, adapters, schemas and fixtures.

Inference first. Training claims and compute accounting are out of scope for now.

Governance

Work in progress. Will be defined as we onboard more teams to the effort and establish how we want to work together.

  • Methodology changes by public RFC
  • Licence: code Apache-2.0; specifications, data, results and reports CC BY 4.0.

Get involved

Open an issue in inference-proofs.

  • Prover teams: your repository, the workloads you support, and who will review your claim card. The adapter guide covers the rest.
  • Relying parties: the thresholds that would let you rely on a proof (latency, cost per token, soundness level, coverage).
  • Cryptographers: the claim cards and the negative controls are where review matters most.
  • Engineers with a GPU: reproduce a run. An unaffiliated reproduction within tolerance is the only thing that moves a number to reproduced.

Popular repositories Loading

  1. inference-proofs inference-proofs Public

    Provable LLM inference measured on the same hardware: overhead against production serving, proving time, proof size and check time, and what one proof establishes.

    Python 14 1

  2. .github .github Public

Repositories

Showing 2 of 2 repositories
  • inference-proofs Public

    Provable LLM inference measured on the same hardware: overhead against production serving, proving time, proof size and check time, and what one proof establishes.

    ai-verification-observatory/inference-proofs's past year of commit activity
    Python 14 Apache-2.0 1 10 0 Updated Nov 20, 2026
  • .github Public
    ai-verification-observatory/.github's past year of commit activity
    0 0 0 0 Updated Nov 19, 2026

People

This organization has no public members. You must be a member to see who’s a part of this organization.

Top languages

Python

Most used topics

Loading…