Introducing Decision Keep: the black box for every AI decision

Every commercial flight carries a black box. When something goes wrong, investigators do not ask the airline to describe the flight — they pull the recorder and listen to what actually happened.

Now think about the AI systems making decisions inside your organisation today: who gets a loan, which claim is flagged, whose résumé is screened out, which transaction is frozen. Those decisions carry real consequences for real people. And almost none of them have a black box.

If an auditor, a regulator, or your own risk committee asked you tomorrow — "what did your AI decide, and can you prove the record hasn't been changed?" — what would you hand over?

For most enterprises, the honest answer is: a story. Logs scattered across vendors. Screenshots. A console someone can edit. That gap — between claiming a decision was made correctly and proving it — is the problem we exist to solve.

Why we built Decision Keep

We came from the side of the room where the questions get asked: audit, risk, compliance, and the engineers who have to defend their systems to people who didn't build them.

We kept seeing the same pattern. Teams had adopted AI fast, governance slow. The models worked, the dashboards looked healthy, and then a regulator or a customer wanted evidence. Not a metric. Not a screenshot. Evidence that a specific decision was made, by a specific model version, at a specific time, and that nobody had quietly rewritten the record afterwards.

That evidence did not exist — because no one had built the recorder. So we built it.

Decision Keep is the black box for every AI decision. Every automated decision your organisation makes is sealed in a tamper-evident record, signed with your own key, and verifiable offline by the people who need to trust it.

Our mission

Make every automated decision provable — so AI can be adopted with confidence, not blind faith.

We believe the biggest barrier to trustworthy AI is not the models. It is the missing evidence trail. Organisations are freezing good AI projects, or shipping them with unmanaged risk, because they cannot answer a basic question: can you prove what happened?

Our job is to make that answer exist by default — for every decision, on every system, in a form an auditor can verify without our involvement.

We are deliberately narrow. We do not build models. We do not compete with your observability stack. We do one thing: produce the independent, forensic witness for your AI decisions, and keep it sovereign to you.

What we believe

A few convictions shape how Decision Keep works:

• The evidence should belong to you. Your keys, your infrastructure, your audit trail. We are the recorder, not the custodian of your truth.
• Trust must be verifiable, not requested. "Trust us" is not a control. Auditors should be able to check a receipt against your published key, offline, with no account.
• Raw personal data should not be the audit trail. We store references and hashes, not copies of people's information — so proving a decision doesn't create a second privacy problem.
• Deletion and defensibility are not opposites. When retention expires, content is cryptographically erased while a signed proof keeps the chain intact. You can satisfy the right to erasure and stay auditable.
• Self-hosted is a feature. For evidentiary integrity, the evidence should never leave infrastructure you control.

How Decision Keep helps your organisation

If you are the person who owns the AI-risk gap, here is what changes when you turn Decision Keep on:

• You can prove every decision. Each one is signed with your Ed25519 key and linked into a hash chain. Change any entry and the break is detectable.
• You can anchor time. An independent RFC 3161 timestamp proves exactly when a decision was signed — turning "we think it was Tuesday" into "signed at 2026-07-16T09:31:02Z, independently witnessed."
• Your auditors verify offline. No login, no trust in us. They check receipts against your published key, in a script or in a regulator's office.
• You satisfy the frameworks at once. The same evidence chain maps to the EU AI Act, GDPR, ISO/IEC 42001, SOC 2 and Australia's APP 1.7–1.9 — one record, many obligations.
• You stay sovereign. Decision Keep runs on your infrastructure. Your keys and evidence never leave it.

Who Decision Keep is for

• Compliance, risk and audit teams who must answer to regulators and the board — and are tired of answering with metaphors.
• Engineering and ML leads who want to ship AI responsibly without bolting a forensic trail onto every service by hand.
• Public-sector and regulated industries — credit, insurance, recruitment, fraud, healthcare triage — where decisions carry legal weight.
• Any organisation adopting agents and autonomous workflows that need a durable, queryable memory of what was decided and why.

A five-minute mental model

Think of each AI decision as a line in a flight recorder:

1. The system decides → Decision Keep captures the model version, inputs (references, not raw PII), output and routing outcome.
2. The record is signed with your key and chained to the previous one.
3. An independent timestamp anchors when.
4. Your auditors verify the whole chain offline, against your published key.

No black box of our own. The evidence is yours, and so is the proof.

Where to start

You do not need to rip out your models. Decision Keep sits beside them and quietly produces the evidence your governance team already owes the regulator.

• See it work: verify a sample receipt with no account.
• Understand the format: read the receipt specification.
• Talk to us: book a demo and we'll map Decision Keep to your automated-decision obligations.

Frequently asked questions

What is Decision Keep? Decision Keep is the independent, tamper-evident record of every automated decision your organisation makes. Each decision is signed with your own key, sealed in a chain that cannot be changed without detection, and verifiable offline by your auditors. It is the black box for your AI decisions — the forensic witness your auditors and regulators can check without trusting us.

Why does the world need a "black box" for AI decisions? Every flight has a black box; most AI decisions have nothing. When an auditor, regulator or your own risk committee asks what an AI decided and whether the record is intact, most organisations can only offer a story. Decision Keep exists so the answer is evidence: a signed, chained, time-anchored record your organisation controls.

What problem does Decision Keep actually solve? The gap between claiming a decision was made correctly and proving it. Logs can be edited, screenshots can be forged, and vendor consoles can disappear. Decision Keep produces evidence an independent party can verify offline, which is what audits, the EU AI Act, GDPR and Australia's APP 1.7–1.9 actually require.

How is Decision Keep different from model monitoring or observability? Observability tells you how a system is behaving now; Decision Keep proves what a specific decision was, when, by which model version, and whether the record is intact. They are complementary — monitoring is operational, the record is juridical.

Do we have to trust Decision Keep with our data? No. Decision Keep is self-hosted and sovereign: it runs on infrastructure you control, your keys never leave your environment, and only content hashes — not raw personal data — are published. Verification is fully trustless; auditors check receipts against your published key with no account and no trust in us.

Who is Decision Keep for? Any organisation that uses AI to make or support decisions with real consequences — credit, insurance, recruitment, fraud, healthcare triage, public services — and the auditors, risk, legal and compliance teams who must answer for those decisions.