Signed, auditable AI

What Is a Tamper-Evident Audit Trail Signed With Post-Quantum Cryptography?

A tamper-evident audit trail is a record of decisions built so that any later alteration is detectable — not prevented, but impossible to hide. When each entry is signed with cryptography, it becomes independently verifiable, not just logged: anyone can check the signature and confirm the record has not changed since it was written. Hibiscus signs these records with ML-DSA, a NIST-standardized post-quantum signature scheme (a provisional patent has been filed). For AI, it means every decision the system makes can be signed and anchored for independent audit — which is exactly what regulated buyers in fintech, healthcare, and federal work need to defend those decisions later.

Definition

Tamper-evident audit trail

A durable record of what a system decided and why, structured and sealed so that any change made after an entry is written can be detected. Its defining property is not that records cannot be touched, but that touching them cannot be concealed — verification either passes clean or fails openly.

Signed audit trail

A tamper-evident trail in which each entry carries a cryptographic signature over its contents. This makes the entry independently verifiable: a third party can recompute the hash and check the signature themselves, without trusting the system that produced it. A signature is the difference between "the log says so" and "here is proof."

Logged vs. signed: why "we have logs" is not enough

Most systems log. Logs are useful for debugging, but as evidence they are weak: an application log can be edited, truncated, reordered, or reconstructed after the fact, and it usually captures only a fraction of the decision context. "Trust me, the log says so" asks an auditor to trust the very party under scrutiny. A signed, tamper-evident trail removes that ask entirely.

 Ordinary logSigned, tamper-evident trail
Editable after the fact?Yes, silentlyYes — but the edit is detectable
Who has to be trusted?The party that owns the logNo one — math, not trust
Independently verifiable?NoYes — recompute the hash, check the signature
Captures full decision context?Usually partialInputs, context, output, all under one signature
Holds up years later?Only if nobody doubts itYes — the signature is the proof

What a signed decision record actually contains

For the trail to be worth anything, each entry has to capture enough to reconstruct the decision, then seal all of it together:

Because the signature covers the whole record, you cannot quietly swap the output, edit an input, or drop an inconvenient piece of context. Change one byte and verification fails. That single property — detectable alteration — is what turns a record into evidence.

Why post-quantum (ML-DSA) specifically

The signing scheme matters because an audit trail is a long-lived asset. A signature you write today may need to hold up in a dispute or audit years from now. Classical signature schemes are vulnerable to future quantum attacks, which would let an adversary forge a signature and retroactively undermine records you thought were settled.

ML-DSA in one paragraph

ML-DSA is the Module-Lattice Digital Signature Algorithm, the signature scheme NIST standardized for the post-quantum era. It rests on lattice problems believed to resist attack even by a large quantum computer. Signing an audit record with ML-DSA means the provenance you establish today stays defensible against tomorrow's cryptanalysis — the whole point of a record meant to survive an audit horizon. Hibiscus built a tamper-evident audit trail on this foundation, and a provisional patent has been filed on the approach.

What this means for AI

An LLM hands you a confident answer whether or not it is correct, and by default leaves no durable record of how it got there. Bolt a signed, tamper-evident trail onto the system and that changes: every decision the model makes can be signed and anchored for independent audit. When an auditor, regulator, or customer asks you to show your work, you can prove exactly which model version answered, what was in the context window, what tools it called, and that none of it was altered afterward.

The order of operations in practice:

  1. Instrument the decision, not just the response. Capture inputs, model and version, context, and tool calls as one structured record at the moment the output is produced.
  2. Sign and anchor each record. Compute an ML-DSA signature over the full record and anchor it so any later modification breaks verification.
  3. Give auditors an independent verifier. Let a third party recompute and check signatures themselves, so trust never has to route through you.
Signing proves what happened — not that the answer was good. A perfectly signed wrong answer is still wrong. That is why a signed trail pairs with an evaluation harness that backtests output against historical ground truth: one proves provenance, the other proves quality. Together you get decisions that are demonstrably correct and independently verifiable. See how signing and evaluation fit together →

Why regulated buyers need this

If you sell AI into regulated markets, "our model is really good" is not something your buyer can act on. Their compliance, risk, and legal teams have to show their work under audit, which means your system has to produce evidence, not just output.

In all three, the buyer is not asking whether your model is clever. They are asking whether they can defend a decision it made, months later, in front of someone with authority. A tamper-evident audit trail is how you say yes.

How this differs from runtime guardrails. Guardrails try to stop a bad output before it happens; a signed audit trail proves what happened after, in a way an auditor can verify without trusting you. They solve different problems, and regulated buyers usually need the proof, not just the filter. Read the full comparison: signed, auditable AI vs. standard LLM guardrails →
Who builds this. Hibiscus Consulting LLC is the software studio of Blake Burnette, a founder-engineer in Cary, North Carolina who ships production AI end to end — model gateways routing tool-using agents, RAG, evaluation harnesses that backtest output against historical ground truth, and a live real-time voice agent you can actually talk to. He also built Nectar, a programming language in Rust that compiles to WebAssembly (public compiler, 2,500+ tests), and the tamper-evident audit trail described here (ML-DSA; provisional patent filed). Blake spent about three years as Director of Engineering at a neobank, running $720M in payments at 100% uptime across 17 engineers and 5 products (Stripe Connect multi-tenant, Plaid ACH, Apple/Google Pay, fraud rules, KYC/KYB); he owned all commits on the Snap! Spend customer-facing React app and wrote core payments code across the platform, and built a React dashboard on a Rails EHR (Medaxion), so the regulated-systems bar is not theoretical here. Hibiscus is an SBIR-eligible small business.

Want AI you can actually prove?

For production AI where every decision is captured, signed with post-quantum cryptography, and independently auditable — and where quality is backtested against ground truth — get in touch.

Email blake@hibiscus.buzz
Founding-engineer, senior/staff/director, or contract. Remote (US) or Triangle-local.

Frequently asked questions

What is a tamper-evident audit trail?
A tamper-evident audit trail is a record of decisions built so that any change made after an entry is written can be detected. Its defining property is not that records cannot be altered, but that alteration cannot be concealed: verification either passes clean or fails openly. It captures the inputs, context, and output of each decision so the decision can be reconstructed and defended later.
How is a signed audit trail different from ordinary logs?
An ordinary log can be edited, truncated, or reconstructed after the fact, and it asks an auditor to trust the party that owns the log. A signed audit trail attaches a cryptographic signature to each entry, so a third party can recompute the hash and verify the signature independently, with no need to trust the system that produced it. That makes each entry independently verifiable rather than merely logged.
Why sign with post-quantum cryptography (ML-DSA)?
Audit records are long-lived and may be challenged years later. Classical signatures are vulnerable to future quantum attacks that could forge a signature and retroactively undermine a record. ML-DSA is the NIST-standardized, lattice-based post-quantum signature scheme, so provenance established today stays defensible against tomorrow's cryptanalysis. Hibiscus built its tamper-evident audit trail on ML-DSA, and a provisional patent has been filed on the approach.
What does a tamper-evident audit trail mean for AI specifically?
It means every decision an AI system makes can be signed and anchored for independent audit. The record captures which model and version ran, the exact context and retrieval results, any tool calls, and the output, all under one signature. When an auditor or regulator asks you to show your work, you can prove what the model saw and that nothing was altered afterward, instead of asserting it.
Isn't a signed answer still just an answer? How do I know it's correct?
Signing proves what happened; it does not prove the answer was good. That is why a signed trail pairs with an evaluation harness that backtests output against historical ground truth, replaying model, prompt, or retrieval changes against past cases with known correct answers and measuring accuracy before shipping. Signing plus evaluation gives you outputs that are demonstrably correct and independently verifiable.