Machine verification infrastructure
attest is independent verification infrastructure for AI and machine activity. Every action your systems take becomes a durable, cryptographically signed record — a signed receipt of what was submitted and accepted, held outside the systems that produced it. attest does not certify objective source truth.
+1,000,000 verified events and counting
01Logs
Your application writes its own logs. Your AI records its own actions. Your agents generate their own audit trails. That makes logs essential for operations — but it also means the system being examined is usually the system providing the evidence.
If that system is compromised, misconfigured, incomplete, or simply wrong, its logs can be wrong too. Records can be changed, deleted, omitted, or never created in the first place.
When a machine action is sent to attest, a separate system receives it, verifies the record, protects against replay, and preserves the resulting evidence independently of the system that produced the action.
The originating system can keep running. Its logs can keep doing what logs do best: debugging, monitoring, and operations. But there is now another record outside that system — one designed for verification rather than self-reporting.
attest cryptographically hashes and signs the evidence it accepts so the integrity of that record can be checked independently of the originating system.
02Why now
Frontier AI is becoming more capable, more autonomous, and harder to verify from self-controlled logs alone. Attest creates an independent evidence layer across machine boundaries.
At the same time, companies are being asked for more accountability: what acted, what happened, when it happened, and what evidence exists to support the answer. Regulatory and governance frameworks are increasingly moving toward auditability, traceability, and evidence — just as autonomous machine activity is accelerating.
That creates a widening gap:
Machines are doing more. Traditional systems are recording themselves. Independent evidence infrastructure has not kept pace.
03How it works
Every machine event successfully submitted to and accepted by attest produces an independently held verification record — outside your infrastructure, outside your cloud, outside of where you’re most vulnerable. Your systems keep running. attest keeps track.
You can retrieve your verification history, confirm any event's status, and export records for auditors, counterparties, or regulators — backed by a verification layer your own team cannot modify.
As your volume grows, your evidence grows with it. Every Verification Pass adds to a cumulative, independently custodied record of what your systems submitted and attest accepted.
Not AI verifying AI — attest uses cryptography and the machine operations already in your stack to create a verifiable record.
An agent calls a tool, a model returns an inference, a workflow completes a step — anywhere in your stack, at any volume.
One authenticated request carries the event to attest with its own identity, from any language or runtime.
The record is canonicalized, hashed, signed, and checked against replay — then persisted outside the system that produced it.
A durable Verification Pass joins your record — retrievable and exportable under the Service retention terms.
04Platform
Verification only matters if the layer holding it is harder to dispute than the systems it records. attest is engineered as durable infrastructure — a small surface with strong guarantees, producing evidence that outlives the moment it was created.
Every accepted event is canonicalized, hashed, and signed. Integrity can be checked independently — without trusting the system that produced the event, or taking attest’s word for it.
Each event carries its own identity. Retries and replays are recognized as duplicates and excluded — one action, one receipt, one charge.
Records are preserved outside your infrastructure and outside your cloud. Your own team cannot rewrite them — which is exactly what makes them worth showing to someone else.
Sandbox and Production are separated end to end, with their own credentials and their own history. Prove the integration before a single Production event is billed.
Environment-scoped API keys are shown once, rotate cleanly, and revoke instantly — credentials that behave the way your security review expects.
Pull verification history by period and environment, as CSV or through the reporting API — shaped for auditors and counterparties, not only dashboards.
One verification layer across the machine economy — the same receipt, whatever the system behind it.
05Pricing
One unique verified event equals one Verification Pass. Duplicates are excluded.
Usage-based, with no minimums. Prove the integration in Sandbox before Production is billed.
A Verification Pass is created when attest successfully verifies one distinct machine action submitted with its own event identity.
If the same event is retried or replayed with the same identity, attest recognizes it as a duplicate and does not create or bill another Verification Pass.
06Docs
Integration is one authenticated request. Send machine activity with your environment credential, watch it verify in Sandbox, then point the same call at Production.
npm install @attestinfra/sdk@0.2.0await attest.verify({
eventType: "ai.agent.tool_call",
source: "research-agent",
eventId: job.id
});
// → verified · signed · custodiedRaw HTTP works too — POST /v1/events with a Bearer credential. Any machine that can make an authenticated HTTPS request can use attest.
const prepared = attest.prepareWitness({
eventType: "transfer.requested",
source: "checkout",
destination: "payments",
action: "transfer",
evidence: { actionId: "transfer_82" }
});
const { result, witness } = await prepared.execute(() => transferFunds());
// → assembly observation · delivery observation · matched0.2.0 is published. Witness records machine observations before and after an operation crosses a system boundary, then reconciles those observations into authoritative pair state. Missing delivery does not establish whether the operation occurred.
07Research
attest publishes the case for independent evidence custody rather than asserting it. Everything below is readable in full and nothing is gated — the standards comment we filed, the paper on what independent evaluation actually requires, and the external research the argument rests on.
Embedded evaluators make judgment independent. They do not make the evidence that judgment rests on independent — and the OpenAI–Hugging Face investigation is the field test.
Every recordkeeping obligation now on the books lands on the same finding: the entity being examined is the entity that holds the evidence.
AI is producing work faster than people can verify it, opening a measurable gap between automation and accountability.
Start verifying
Start in Sandbox in minutes. Your first verified event becomes your first receipt — and the record only grows from there.