The accountability layer for autonomous AI decisions.

Sovereign creates cryptographically sealed, independently verifiable records that bind a named, authorized human to every consequential AI decision.

Across every sector, courts and regulators are asking the same question: Who decided?

When an AI-influenced decision is challenged, the institution cannot show who owned it. Courts are now assigning liability for that gap.

Every case was missing the same record: which authorized human owned the decision, verifiable later, by anyone. That record is what Sovereign produces.

A log records the decision. Sovereign proves who stood behind it.

Most AI systems already keep logs. It can be changed, it sits inside the application, and an outside reviewer has no independent way to establish what happened. Sovereign creates a separate record, signed by the named human who approved the decision, that can be verified without relying on the system itself.

One decision. One human sign-off. Independently verifiable.

The approver reviews AI recommendation and evidence, records their rationale, and signs with a passkey on their own device. Sovereign then seals the decision and records its proof in a transparency log. Anyone can later verify what was signed and detect any change to the record, without trusting the originating application, dashboard, or audit log.

approver dashboard

sovereign/queue audit··

sovereign sign-off

Decisions awaiting your review

My reviewsUpcomingWaiting on othersClosed

adverse_action.credit_denial · case 2216-084

needs your sign-off

model recommends: deny · policy 2026.05.12 · evidence sha256:b8d1…42c7

View the recommendation & evidence ›

Your rationale (required) 98 / 600

Income verification failed under policy 4.2. The model's evidence is consistent; I uphold the denial.

⚠Your rationale is signed and recorded in the public transparency log — don't include personal or health information (PII/PHI). Keep it about the decision, not the data.

By signing, you confirm that you personally reviewed this evidence and made this decision.

Sign with passkey

audit verification

sovereign/audit queue··

source: transparency log

Audit verification

Enter a ledger UUID to verify its cryptographic entry in the transparency log.

ledger uuid

01J9R2X9F0ENVZ7

Verify record

raw log entry (json)

Results will display here...

verify this record yourself — offline, no server trust. the cli verifier (verify-ssp.py · trust-public.pem) re-checks everything; the trust key is supplied separately, so a bundle can never self-certify.

Why the proof holds.

Five cryptographic steps turn a decision into evidence anyone can check without trusting Sovereign. Each is a public standard with public verification tooling.

Two signatures. One the institution cannot forge, one the person cannot deny.

The approver signs on their own device; a customer-held kms key then wraps that signature so it cannot be detached from the claim or its anchors. Sovereign is never in the trust loop.

A logged reviewer field can be written by anyone with database access. A webauthn signature can be produced by exactly one device, held by exactly one person — the difference a court cares about.

Runs in your cloud. You hold the keys.

Deploy from the marketplace into your own account: customer kms, customer storage, your domain as the webauthn relying party. Sovereign code never sees private key material.

For decisions where a human owns the call.

The pattern is the same everywhere: a model recommends, a person decides, and someone will later ask who. Built first for EU, Swiss and UK institutions running adverse-action workflows.