Put a Tamper-Evident Receipt on Your AI Design Demo Before You Demo It

Description and use

Implemented a tamper‑evident receipt system for AI design demos that records model ID, prompt and output hashes, rejected alternatives, stop conditions, and a hash chain, enabling reviewers to verify integrity before approval.

Problem

Product teams present AI-generated designs without verifiable provenance, making it hard to prove which model produced the output, what alternatives were rejected, and whether the output was altered.

Outcome

The receipt system surfaced missing model IDs and hash mismatches during a sandbox demo, causing the reviewer to halt approval. Manual edits to previous entries were detected, demonstrating tamper evidence.

Workflow

• 1. Capture agent decision data (modelId, prompt, selected, rejected, stopCondition).
• 2. Compute SHA-256 hashes for prompt, selected, rejected items, and previous receipt entry.
• 3. Append new receipt entry to receipts.jsonl with timestamp, hashes, and previousHash.
• 4. Human reviewer checks current and previous hashes; if chain mismatched or required fields missing, halt approval.
• 5. If all checks pass, approve and proceed to next demo step.

Evidence

The receipt made missing fields visible in my sandbox, and the reviewer stopped when modelId was missing.

Source

I also ran a small manual edit: I changed an old line. The next receipt still acted as if it was new, but the reviewer could see previousHash did not match.

Source