The era of blind trust in AI agent outputs is over, and the mechanism to enforce it is deceptively simple. Zambo has released a granular breakdown of how its AI agent receipts are verified, moving beyond vague promises of transparency to a concrete, cryptographic reality. The core of the system relies on recomputing the SHA-256 digest of the exact saved result and comparing it against the digest recorded at the moment of execution. If the hashes match, the record is intact; if they differ, the agent or the platform lied, and the system flags it immediately.

The Four-Step Verification Protocol

The process begins by opening a public receipt link, such as zambo.dev/run/[id], which displays the tool used, inputs, and the saved result without requiring an account. The second step involves recomputing the digest of the 'canonical bytes' of that result. This can be done via the browser verifier at zambo.dev/verify/ or locally using the aer1-verify PyPI package or the aer-1 npm package, ensuring no reliance on the platform's own API. The third step is the comparison: a match yields 'Verified,' while a mismatch yields 'Verification failed,' explicitly indicating the record was altered post-execution. Finally, the fourth step walks the record through its larger structures, checking both the sequence of receipts for a job and any external anchors, ensuring a gap in the record is visible rather than silent.

Independence Through Open Standards

What separates this from a black-box platform feature is its foundation in the open AER-1 format, specifically IETF Internet-Draft draft-zambo-aer1-26. The specification is frozen, meaning the target a verifier builds against does not shift. To prove this, Zambo points to a public conformance corpus at zambo.dev/aer1/corpus.json, containing 126 vectors and 129 checks. The critical proof of independence lies in the fact that one verifier was written solely from the draft text, without seeing the reference code, and still agreed with the official kit on 119 out of 119 base vector checks. Two pieces of software sharing no code and agreeing on every check is the only metric that matters in a trustless system.

Scale and Limitations of Cryptographic Proof

As of November 1, 2026, Zambo has issued 92,515 public receipts across 80,599 tool calls, with 2,442 generated in a single day. Every one of these is free for any stranger to check using the four steps outlined. However, the guide is honest about what this verification does not prove: it confirms the record was not changed, not that the result was correct. A verified receipt can still contain a bad price or a weak report. The mechanical check removes the argument about tampering, leaving the judgment of quality to the human reader.

Key Takeaways

  • Verification is mechanical: It relies on SHA-256 digest comparison, not trust in the agent's narrative.
  • Open standards: The AER-1 format is frozen (draft-zambo-aer1-26) with a public conformance corpus to ensure independent verifier compatibility.
  • Independence proven: A third-party verifier written without reference code matched the official kit on 119/119 checks.
  • Scale: Over 92,000 public receipts are currently issued and verifiable for free by anyone.
  • Limitations: A verified receipt proves integrity (no tampering), not accuracy (correctness of the output).

The Bottom Line

Zambo’s approach finally shifts the burden of proof from the agent’s confidence to mathematical certainty. By making verification a free, mechanical, and open standard, it eliminates the 'trust me' economy that has plagued AI agents, forcing platforms to expose their records to independent scrutiny.