What Is a Cryptographic Decision Record?

A cryptographic decision record is a signed, tamper-evident receipt created at the exact moment an AI model produces an output. It contains the model version ha

A Receipt for Every AI Decision

A cryptographic decision record is a digitally signed receipt that proves exactly what an AI model decided at a specific moment in time. Unlike a log entry or database row, a cryptographic decision record cannot be altered, deleted, or backdated after creation. The mathematical properties of the cryptographic signature make any tampering immediately detectable by any third party with access to the verification key.

What Is Inside a Decision Record

Each InferTrust™ decision record contains seven fields that together constitute a complete evidentiary snapshot of an AI inference:

• Model Version ID (MID): A cryptographic hash of the exact model binary that produced the inference. Not a version label or tag, but a hash of the binary itself. If the model changes by a single byte, the MID changes.
• Confidence Score (CS): The model's self-reported confidence at inference time. Used to determine whether the decision can proceed autonomously or must be escalated to human review.
• Policy Version Hash (PVH): A hash of the organizational policy ruleset active at the time of the decision, including confidence thresholds, allowed actions, and escalation criteria.
• Input Feature Hash (IFH): A hash of the input data features the model evaluated. This proves what the model saw without storing the raw data, which is critical for privacy-sensitive environments like healthcare.
• Timestamp (TS): A cryptographically bound timestamp at the moment of inference, proving when the decision was made.
• Decision Action (DA): The specific action the model recommended or executed, such as APPROVE, DENY, ESCALATE, or FLAG.
• Sequence ID (SEQ): A strictly incrementing counter baked into each record. If anyone deletes or reorders records, the gap in the sequence is visible, proving the log is complete.

How It Differs from a Log Entry

A log entry is a row in a database. It can be edited, deleted, or overwritten by anyone with database access (or by a bug, or by a compromised system). A log entry has no mathematical proof that it was created at the time it claims, or that its contents have not been changed since creation. A cryptographic decision record has both. The signature binds the content to a specific moment and a specific device. Any modification breaks the signature and is immediately detectable.

When Is the Record Created

The record is created and signed at the point of inference, inside the inference event, before any network transmission, before any UI rendering, before any application layer processes the output. This is the critical architectural difference. There is no gap between the decision and the record. The signed receipt is the first artifact that exists after the model computes its result. This eliminates the window during which conventional logging systems are vulnerable to modification.

Who Can Verify a Decision Record

Any party with access to the verification key can independently confirm that a decision record is authentic and unaltered. This includes internal auditors, external regulators, opposing counsel in litigation, and automated compliance systems. The verification does not require trust in the organization that created the record. The mathematics provide the assurance directly.