ArkForge Trust Layer: using Rekor as public transparency anchor for AI agent certification

[desiorac]

What is ArkForge Trust Layer?

ArkForge Trust Layer is an open-source certifying proxy for AI agents. It sits between any AI model (Claude, GPT, Mistral, open-source models) and external APIs, creating cryptographic proofs of every action an AI agent performs.

How we use Rekor

Every proof generated by ArkForge Trust Layer is anchored to the Sigstore Rekor public transparency log using hashedrekord v0.0.1 entries (ECDSA P-256 + SHA-256).

Concretely:

• Each certified API call produces a proof chain (input hash → output hash → chain hash)
• The chain hash is submitted to Rekor via the public instance API
• The resulting logIndex and uuid are included in the proof returned to the caller
• Anyone can independently verify the proof via search.sigstore.dev

This gives our users public, immutable, third-party evidence that a specific AI action happened at a specific time — without trusting ArkForge itself.

Why Rekor?

We evaluated several transparency log options. Rekor won because:

1. Public and free — no vendor lock-in, no cost barrier
2. Append-only with inclusion proofs — exactly what we need for tamper evidence
3. Well-maintained with a 99.5% SLO on the public instance
4. Standard format (hashedrekord) that does not require Fulcio certificates

Architecture snippet

# Simplified flow in our proxy
proof_chain_hash = sha256(input_hash + output_hash + timestamp)
rekor_entry = submit_to_rekor(proof_chain_hash)  # hashedrekord v0.0.1
proof["rekor"] = {
    "log_index": rekor_entry["logIndex"],
    "verify_url": f"https://search.sigstore.dev/?logIndex={rekor_entry['logIndex']}"
}

Links

• Trust Layer API: https://trust.arkforge.tech
• GitHub: https://github.com/ark-forge/arkforge-trust-layer
• Rekor integration code: trust_layer/rekor.py

We are happy to answer any questions about this use case. Rekor has been a great fit for bringing public verifiability to AI agent actions.