01
Hash both sides
The exact input and output are hashed separately with SHA-256, preserving selective disclosure later.
input_hash · output_hashSYSTEM 03AI ATTESTATION
Turn every AI response into evidence.
Capture the exact prompt, response, model, and moment. Invoance hashes the exchange, signs the record, and gives anyone a way to verify it without trusting a screenshot or an editable log.
One call after generationNo proxy in your model pathPortable verification
ATTESTATION / LIVE RECORDSIGNED
INPUT · SHA-256
Should this commercial claim be escalated for manual review?
7db0a6f4 3ef98c21 b12eaa09 941cf8d2
OUTPUT · SHA-256
Escalate. Confidence 0.93. Policy threshold 0.81.
21be74f8 9a67c011 f4ac1290 a50d6e91
MODELgpt-4.1
ANCHORED14:32:07.123Z
PROOF IDatt_01JQ7F…P8E
ED25519 SIGNATURE
6fB3kYw9n2M4dQ7t…eA9101
CAPTUREExact exchange
02
HASHSHA-256 digest
03
SIGNEd25519 proof
04
VERIFYIndependent check
01PLACEMENT
Nothing changes upstream.
Your model still receives the prompt and returns the response exactly as it does today. Attestation runs after generation, on a copy of the exchange—not between your application and the AI.
YOUR SYSTEMApplication
INFERENCEAI model
DELIVERYResponse
ASYNCHRONOUS COPY
POST-GENERATIONInvoance anchors the exchange
POST /v1/ai/attestationsNo proxy. No model change. No new inference dependency.One API call after the answer leaves your model.
02EXCHANGE
One request in. One proof out.
Send the source material required to reconstruct the moment. Receive hashes, a signature, an immutable identifier, and a verification route.
Python REQUEST
invoance-pythonproof = await client.attestations.ingest(
attestation_type="output",
input="Review this claim...",
output="Escalate for review...",
model_provider="openai",
model_name="gpt-4.1",
model_version="2026-04-14",
)Python RESPONSE
AttestationIngestResponseAttestationIngestResponse(
attestation_id="att_01JQ7F...P8E",
created_at="2026-07-21T14:32:07Z",
input_hash="7db0a6f43ef9...",
output_hash="21be74f89a67...",
payload_hash="f93c0d51c240...",
status="accepted",
)03PROOF PIPELINE
Five deterministic operations.
Every attestation follows the same inspectable sequence. The result is not a claim that the output was good—it is proof of exactly what happened.
02
Build one record
Model, version, parameters, metadata, hashes, and timestamp become one canonical payload.
canonical_attestation03
Sign the bundle
An Ed25519 signature binds the complete record to your organization and exposes any later change.
signature · public_key04
Commit append-only
The signed bytes are stored as a permanent record. The system supports create and verify, never rewrite.
status: anchored05
Make it verifiable
A proof URL and portable verification material let another party check the record independently.
verify: offline-ready04VERIFICATION
Trust the math, not the database.
A verifier recomputes the disclosed hashes, rebuilds the attestation digest, and checks the Ed25519 signature against the published key. The proof still works when Invoance is not in the room.
Selective disclosureVerify input or output independently.
Third-party readyAuditors do not need your dashboard login.
LOCAL VERIFIERNO NETWORK REQUIRED
01
Load signed attestationDONE02
Recompute input SHA-256MATCH03
Recompute output SHA-256MATCH04
Rebuild canonical digestMATCH05
Verify Ed25519 signatureVALIDVERIFICATION RESULTRecord is authentic and unchanged
05SCOPE
Proof of the event. Not a verdict.
Attestation establishes technical facts about an AI interaction. It deliberately does not make claims about whether the model was correct, fair, or legally compliant.
WHAT IT PROVES
+The exact input the model received
+The exact output the model produced
+The model and version used
+When the generation was anchored
+Whether any signed field changed
WHAT IT DOES NOT PROVE
−That an output is factually correct
−That a model is unbiased
−That the original prompt was appropriate
−That a human used the output correctly
−Automatic legal admissibility
06CONTROL EVIDENCE
Built for the questions that arrive later.
Regulations and assurance frameworks ask for traceability, integrity controls, and evidence. Attestation supplies the technical record those processes can inspect.
ART. 12 · 13EU AI ACT
Traceable records for high-risk AI systems
MAP · MEASURENIST AI RMF
Evidence of system behavior and controls
CC6.1 · CC7.2SOC 2
Integrity and monitoring evidence
AI MANAGEMENTISO 42001
Operational accountability and auditability
Technical infrastructure, not legal advice.Requirements vary by system, sector, and jurisdiction. Invoance preserves verifiable evidence; your compliance program decides how that evidence is used.
READYSHIP THE PROOF
Your model already speaks. Put it on the record.
Add one call after generation and turn every important AI exchange into evidence that can survive an audit, dispute, or handoff.
How AI Attestation Works, Invoance
Technical walkthrough of how Invoance AI Attestation creates tamper-evident records of AI inputs, outputs, and model metadata at the moment of generation. Anchor AI payloads with SHA-256 hashing and Ed25519 signing for permanent, independently verifiable proof of what any AI model produced and when.