CarveTrace
Cryptographic AI Act evidence — your auditor can re-verify without trusting us.
Aryamind SARL · carvetrace.com · 2026
Pitch deck
The CarveTrace pitch in ten slides
Designed for a 5–7 minute IAPP-floor pitch, expanding to 15 minutes with the demo slide opened in the browser. Hit Ctrl+P and save as PDF to email this deck ; the print stylesheet renders one slide per page.
Source : sales/PITCH_DECK.md on GitHub also carries speaker notes for each slide.
The deadline
On August 2, 2026, EU AI Act Article 12 (record-keeping) + Article 14 (human oversight) bind for every high-risk AI system.
National regulators get investigative powers. Fines reach €15M / 3% global turnover. Up to €35M for the worst categories.
Most enterprises have a governance program. Few have evidence a third party can verify.
The deeper deadline
December 9, 2026 — Product Liability Directive 2024/2853 extends strict liability and presumption of defect to AI software.
When a claimant alleges harm, the burden of disclosure shifts to the producer.
If your evidence requires you to be alive to defend it, your evidence is not evidence.
What every AI governance platform ships today
"Trust our database. We'll show your auditor a dashboard."
CredoAI · Holistic AI · ModelOp · Saidot · OneTrust · IBM watsonx.governance · ServiceNow · Vanta · Drata.
All of them : if the vendor disappears, the regulator subpoenas them, or their tenant database leaks — your evidence goes with it.
Two properties no competitor ships
1. Re-verifiable without trusting CarveTrace. WASM verifier, Apache 2.0. Your auditor opens verify.carvetrace.com, drops your evidence bundle, gets VERIFIED in seconds. No account.
2. Article 12 ↔ Article 14 bound at write-time. Every AI inference is cryptographically bound to the human-oversight event that reviewed it. A single signed atom.
Show, don’t tell
A real CarveTrace bundle is at verify.carvetrace.com right now.
Seven AI screening decisions. Five working days, three reviewers. Two overrides, two escalates, three approves — including one decision where the AI said "poor fit (0.34)" and a human overrode with the rationale "AI scored 'poor fit' off the 'Analyst II' job title alone ; actual responsibilities match a senior IC role."
Click it. Drag the bundle. Read VERIFIED. Flip one byte. Watch FAILED. Same verifier your regulator will use. Same verdict you will see.
Pricing
Per high-risk AI system × decision volume — the unit the regulation regulates. Self-hosted across every tier. EUR, excluding VAT.
Verifier €0/yr free, forever, no account Starter €15K/yr 1 system, ≤500K decisions Growth €45K/yr 5 systems, ≤5M decisions, SSO, DPA, EU residency Enterprise from €90K unlimited, ISO 42001 pack, named CSM, DORA fit
2-yr commit : −10%. 3-yr commit : −15%.
What "high-risk AI system" means here
One production AI workflow you would list under Article 16 as a deployer or provider :
· one résumé-screener · one credit-decisioning model · one biometric matcher · one tariff-eligibility classifier · one triage-prioritization model
→ one signed CarveTrace chain → one quarterly evidence bundle
60-day pilot
Week 1 — SDK drops into your existing code. No model change. No data leaves your perimeter.
Weeks 2–7 — Real chain emitting. One quarter's worth of evidence accumulates. Direct engineering access.
Week 8 — Hand a bundle to your internal auditor. They verify it in their browser. They sign off.
Modest pilot fee, credits back against your first annual subscription if you convert.
The ask
We are looking for 5 design-partner customers across EU banking, insurance, HR-tech, and public sector — each with one high-risk workflow live by Aug 2.
If that's you : · Demo today, 30 minutes — at any open table, or virtually next week. · Pilot scope conversation this month. · Bundle in your auditor's hands by September.
contact@carvetrace.com verify.carvetrace.com (try it now) carvetrace.com/why (the one-pager)