Independent evidence for enterprise AI
Prove what your AI agents did.
Each step your agents record, signed and sealed to an independent public record. Evidence that auditors, regulators, insurers and counterparties can verify themselves, offline, without trusting you or us.
Your sensitive data never leaves your system
- model
- credit-decision-model v4.2.1
- digest
- sha256:7d86c1a4e2f09b3d…91b0
- signed
- ed25519 · signing key 4f9a…c2e1
- witnessed
- 2026-08-07 14:02 UTC
- sealed to
- an independent public record
Integrations
One call per step. Evidence from the first one.
await anchorer.anchor({ eventType, content })SDKs for TypeScript and Python. About ten minutes from install to a pack anyone can verify: read the quickstart.
Why we exist
As agents act, the cost of trust is exploding.
AI agents now make consequential decisions and take real actions; approving credit, pricing risk, moving money, touching data. Every action creates a question someone will later ask: can you prove what your system actually did? Today that's expensive and complex; manual evidence collection, audits and questionnaires. Slow processes create higher premiums and deals that die in security review. That is the rising cost of trust, and it climbs with every system and agent you deploy.
Sanning compresses the rent.
We make the truth of what your systems did mathematical and provable. You can give the market, regulators and auditors independent proof in seconds, instead of paying, forever, to be believed.
Product
Coverage in one line. Proof on demand.
Independent proof, not your own logs.
01
02
03
04
See it verify
Run 9f3c21b8 · underwriting-triage/4.2.1
Records
Content-blind
raw data stays in your systems
only signed envelopes leave
sealed to an independent public record
Hand-over
Verified offline
Verified offline, with an MIT-licensed kernel and no call to Sanning.
Simulated run · illustrative records
Between agents
When both agents record with Sanning, both sides of a hand-off are on the record.
Each hand-off is recorded twice: the sending agent and the receiving agent each sign it with their own key and commit to the same message. Put every party's pack into one case reading and you see whether the message the second agent received is exactly the one the first sent, even across two companies. Read offline, with the MIT-licensed kernel.
Works over any transport. Record a hand-off.
Who it's for
Built for decisions that must be defensible for years.
Banks · Insurers · Healthcare · Auditors & underwriters · AI platforms
Financial services, insurance and healthcare. Anywhere an AI decision has to hold up to an auditor, a regulator, insurer or a court, long after it was made. We start where the stakes are highest: regulated underwriting and model risk.
Risk & controls
Hand auditors independent evidence, not internal logs.
A pack an auditor can verify, in one command.
AI & ML engineering
Control and audit without redesigning the stack.
A few lines, evidence from day one.
Auditors, insurers & counterparties
Rely on proof you didn't create.
Verified independently, by anyone.
Trust
Evidence for the records these rules ask you to keep.
Laws, supervisory expectations and standards like these ask for records. Sanning is how you show the records you kept were not altered. Your auditor decides what that satisfies.
Built on open standards anyone can verify.
RFC 8785 JCS · SHA-256 · Ed25519 · RFC 9162 · TypeScript and Python kernels, MIT-licensed
Prove what your AI systems and agents did.
See it verify a real decision in under a minute.