EU AI Act Article 12 enforcement begins August 2, 2026 — 92 days away

AI Decision Audit Trail: Requirements and Implementation

Under the EU AI Act, every high-risk AI system must maintain a verifiable audit trail of automated decisions. This is not optional — enforcement begins August 2, 2026.

What is an AI Decision Audit Trail?

An AI decision audit trail is a tamper-evident, chronological record of every automated decision made by an AI system — including the inputs that led to the decision, the policy or model that evaluated it, the verdict reached, and the timestamp of inscription.

Unlike a simple log file, a proper audit trail must be cryptographically verifiable — meaning any tampering is immediately detectable. This is what distinguishes compliance-grade audit trails from standard application logging.

What EU AI Act Article 12 Requires

Log every automated decision
Every decision made by a high-risk AI system must be recorded — not sampled, not summarised.
Record the inputs that led to the decision
The data and parameters used to reach each decision must be stored alongside the verdict.
Maintain logs for a minimum of 6 months
Records must be retained and available for inspection for at least 6 months after the decision.
Ensure logs are available to authorities on request
National supervisory authorities must be able to access and verify the audit trail.
Provide a human oversight mechanism
The system must allow human review and override of automated decisions — not just logging.

How Krapheno Implements the Audit Trail

Requirement
Automated decision logging
Krapheno Implementation
SmritiTree append-only ledger — every decision inscribed at the moment it occurs, not in batch.
Requirement
Input recording
Krapheno Implementation
Full decision payload stored with each inscription — sample size, confidence, segment concentration, policy version.
Requirement
6-month retention
Krapheno Implementation
Hash-chained ledger is permanent and tamper-evident — records cannot be deleted without breaking the chain.
Requirement
Authority access
Krapheno Implementation
Public trace endpoint per decision — any record verifiable independently via cryptographic proof.
Requirement
Human oversight
Krapheno Implementation
ESCALATE verdict triggers human review — approve or reject directly from the governance portal.

Technical Architecture

Krapheno uses a SHA-256 hash-chained ledger called SmritiTree. Each record contains the decision payload, the policy verdict, and a cryptographic link to the previous record. Any attempt to modify or delete a record breaks the chain — immediately detectable by any independent verifier.

Decision inscribed
→ payload + verdict + timestamp
→ SHA-256(prev_hash + payload) = record_hash
→ stored append-only in SmritiTree
→ verifiable via GET /v1/decisions/verify

See a live governed decision

Krapheno has been logging real AI decisions since day one. Every record cryptographically signed and independently verifiable.

See live demo →EU AI Act Article 12 guide
Governed by Krapheno SmritiTree · Cryptographic audit trail enabled