Cryptographic execution receipts for AI agents

The execution receipt layer for AI agents.

Cryptographic proof of every skill an agent runs — audit-grade, regulator-ready, signed before execution.

EXECUTION RECEIPT #AXM-4F8C-9A2E
skill finance.reconcile.v2
invoked 2026-05-27 04:18:32 UTC
inputs_hash sha256:e3b0c44298fc1c14…
output_hash sha256:a1b2c33d9e7f01a2…
attestation ✓ VERIFIED
algorithm HMAC-SHA256 + timing-safe verify
The problem

AI agents are running in production. No one can audit what they did.

Regulators are catching up fast. The infrastructure hasn't.

11.9%
of MCP skills audited as malicious
Koi Security ClawHub audit: 341 of 2,857 registered skills flagged. Antiy CERT identified 1,184+ additional variants. Koi Security
Aug 2, 2026
EU AI Act enforcement begins
Fines up to 7% of global annual turnover for non-compliant AI systems. Requires auditable, identifiable agent execution records. EU AI Act
10,000+
MCP servers with zero provenance tracking
The Model Context Protocol ecosystem has grown to over 10,000 registered servers. None ship cryptographic execution receipts by default.
How it works

Three steps. One tamper-evident receipt.

Axiom intercepts skill invocations, generates a cryptographic attestation before execution, and makes verification available on demand.

01

Invoke skill

Your agent calls a skill via the Axiom receipt layer. Inputs are hashed with SHA-256 before the skill runs — not after.

POST /api/receipts/invoke
{
  "skill_name": "echo",
  "inputs": {
    "message": "reconcile Q1"
  }
}
02

Receipt generated

Axiom generates a signed receipt — HMAC-SHA256 signature over the skill name, timestamp, and input/output hashes. Stored immutably.

Receipt response
{
  "receipt_id": "AXM-4F8C",
  "inputs_hash": "sha256:e3b0…",
  "output_hash": "sha256:a1b2…",
  "signature": "hmac:f7c3…",
  "status": "VERIFIED"
}
03

Verify on demand

Compliance team or regulator requests verification. Axiom uses timing-safe HMAC comparison — mathematically provable, not just a log entry.

GET /api/receipts/:id
{
  "verified": true,
  "receipt": { … },
  "verification_status":
    "valid"
}
Don't trust the vendor — audit the verifier yourself
MIT-licensed, dependency-free. Node.js and Python. The same crypto that runs server-side.
Open-source verifier → See how it works →
Live demo

A real receipt. Generated against a live MCP server. Right now.

This is not a mockup. The demo calls a production MCP server, generates a cryptographic receipt, and returns a verifiable signature — all in under 200ms.

axiom — live receipt terminal
$ curl -X POST https://axiom-ai-19.polsia.app/api/receipts/invoke \
-H "Content-Type: application/json" \
-d '{"skill_name":"echo","inputs":{"message":"audit test"}}'
→ receipt_id: AXM-7D2B-FA91
→ inputs_hash: sha256:4d4a…
→ signature: hmac:9f7e…
→ status: VERIFIED
Open interactive demo No signup required. Generates real receipts.
Axiom vs. everything else

Logs tell a story. Receipts are proof.

Logging and monitoring tools record what happened after the fact. Axiom attests to what was about to happen — before the skill runs. That distinction is the entire compliance argument.

Capability Axiom Log / Monitoring Observability Platform
Pre-execution attestation Signed before skill runs Post-hoc narrative Post-hoc narrative
Cryptographic proof HMAC-SHA256 Plain text logs Plain text traces
Tamper-evident receipts Immutable, signed Mutable log files ~ Read-only, not signed
Regulator-ready export Structured, verifiable ~ Requires manual curation ~ Requires manual curation
EU AI Act compliance posture Built for it Not designed for it Not designed for it
Input / output hash verification SHA-256 both directions No No
Open-source independent verifier MIT, zero deps No No
Full comparison vs Meridian, MintMCP, Vanta, JFrog →
ROI Calculator

Make the business case in 60 seconds

Your CFO wants a number. EU AI Act fine exposure vs. Axiom annual cost — calculated live, emailed as a citable PDF.

Your inputs
Annual global revenue $500M
$10M$50B
AI agents in production 10
1500
EU AI Act max fine exposure Art. 99 ↗
7% of global annual turnover — worst-case prohibited-practice violation¹
Audit prep cost saved / yr Industry survey²
SOC 2 / HIPAA prep costs Axiom automates away — conservative range
Axiom estimated annual cost ³
$2,000/mo platform fee + $0.01/attestation event (agents × 500 events/mo est.)
coverage ratio — fine exposure ÷ Axiom annual cost
Email me this analysis as a PDF
Get a branded, shareable PDF with footnoted sources — ready to send to your CFO.

Sources

¹ EU AI Act Article 99(4) — maximum administrative fine for prohibited AI practice violations: 7% of global annual turnover or €35M, whichever is higher. Article 99(3) sets 3% for other high-risk violations. This calculator uses 7% as the upper planning bound. Full text (EUR-Lex).

² Audit preparation ranges ($40K–$120K / year) based on SANS 2023 SOC 2 Cost Survey and HIMSS 2023 Cybersecurity Survey. Financial services and healthcare organizations typically fall in the upper half of this range due to regulatory density. Actual savings vary by maturity.

³ Axiom pricing estimate uses $2,000/month base platform fee plus $0.01 per attestation event. Agent event volume estimated at 500 events per agent per month for planning purposes. Actual cost depends on negotiated pilot terms.

Want the full standalone calculator? Open /roi ↗

Why this founder

Built by someone who had to answer to a regulator.

JM
Joseph McCann
Founder, Axiom

20+ years as CFO and COO across nonprofits, education, and operating companies — running Federal Single Audits, GAAP compliance, ITGC controls, and multi-entity treasury across the US, UK, France, and Spain.

Spun out a $12M nonprofit from the Tides Foundation — built the entire financial, HR, and governance infrastructure from scratch, independently, on time. No team. No margin for error.

Before Axiom: built AI.FO — a production CFO-reasoning system with 2,000+ deterministic assertions, a sealed-snapshot architecture, and an audit-independent test harness. Every output provably derived. Every number traceable to its source.

That background made one thing obvious: AI agents running in regulated environments are a compliance disaster waiting to happen. Axiom is the fix.

NYU LSE HEC Paris MBA AI.FO founder Federal Single Audits
Pilot program

Request a pilot

We're opening Axiom to a small cohort of compliance-forward teams. Tell us about your use case and we'll reach out directly.

No spam. No sales calls unless you want one. We'll reply directly.

As featured by the founder → Read: Sealed Snapshots — Audit-Grade AI Architecture