every model decision, prompt, and policy check, written to a linked ledger. tamper the data, the chain breaks. pull evidence for any decision in seconds.
Every AI action gets a unique fingerprint computed from its full payload. Change one byte — even a whitespace — and the fingerprint diverges. You will know, and so will your auditor.
Each record carries the fingerprint of the one before it. Altering a record in the middle of the chain breaks every record after. There is no place to hide a rewrite.
Pull evidence for any AI decision in milliseconds. Hand regulators the full sequence: the prompt, the policy, the reviewer, the output — stamped, linked, and independently verifiable.
Every AI action, human approval, and automated decision is posted to the SYEN Audit API. One record captures identity, policy, data lineage, execution, review, outcome, and cryptographic proof.
{
"event_class": "EXECUTION",
"event_type": "credit.decision.made",
"stream_id": "tenant-demo:prod-us-east-1:credit-decisions",
"payload": {
"applicant_id": "app-991",
"decision": "APPROVED",
"score": 740
},
"policy_context": {
"policy_id": "policy-credit-v3",
"framework_name": "SR_11_7",
"requirement_id": "MRM-GOVERN-4.2",
"policy_effect": "ALLOW"
},
"data_lineage": {
"data_asset_id": "dataset-credit-bureau-2026-q1",
"data_source_system": "snowflake",
"contains_pii": true,
"data_classification": "confidential"
},
"ai_execution_context": {
"model_provider": "anthropic",
"model_name": "claude-sonnet-4-6",
"prompt_hash": "sha256:a3f8c2d1e4b7...",
"inference_latency_ms": 340
},
"guardrail_context": {
"kill_switch_result": "PASS",
"approval_gate_result": "APPROVED",
"risk_score": 0.12
},
"human_review_context": {
"human_review_required": true,
"review_decision": "APPROVE",
"final_approver_id": "user-analyst-007",
"reviewer_role": "credit_analyst"
},
"outcome_context": {
"decision_result": "APPROVED",
"actual_action_taken": "credit.line.issued",
"financial_impact_usd": 25000.00
}
}// awaiting inputEach record is signed with Ed25519 using the customer's own KMS key, then chained to the preceding record. Any alteration breaks the chain and is immediately detectable.
SHA-256→0x7f4a2c1d9e8b…sha256:0000000000000000…genesis anchorthissha256:7f4a2c1d9e8b3f6a2e5d8c1b9a4f7e30…base64:mE9xK2pL8nQ4rT7wY1vB7sDqFzH2jRp…Signing key: alias/syen-audit (customer-controlled). SYEN infrastructure never holds the private key.Any party can verify the chain independently using standard OpenSSL tooling. No dependency on SYEN infrastructure required.
// stream_id is pre-filled from step 1// awaiting inputWhen a regulator or auditor requests proof of how this AI decision was made, this is the answer. Cryptographically signed, timestamp-anchored, independently verifiable.
// event_id is pre-filled from step 1// awaiting inputLog every credit, fraud, and AML decision with the policy, reviewer, and feature inputs attached. Examiners get a cryptographically verifiable trail, not a CSV.
Shopping agents, bill-pay, procurement. Each agent step — plan, tool call, observation, approval — is fingerprinted and linked. Replay the full trajectory; prove it wasn't edited.
Capture the alert, the model, the analyst review, and the outcome. Every chain of custody is replayable and independently verifiable by the investigator's regulator.
FOIA, GDPR Art. 22, EU AI Act — when a citizen asks why a decision was made, pull the full chain of events with one query and a signed proof.