The Zero-Latency Floor.

It is a permanent, tamper-proof log of a single decision made by an AI system. The record captures who was responsible, what the system was trying to accomplish, and the full reasoning chain it used to reach its conclusion. This record is sealed at the exact moment the decision fires, not reconstructed after the fact. Think of it the way you would think of a flight data recorder, a medical chart, or a signed audit trail. Every profession that earned public trust did it by keeping the record. AI is simply the newest profession that needs one.

Because the board is already liable. Directors owe a duty of care under corporate law in virtually every jurisdiction. When AI systems make decisions on behalf of the corporation, that duty does not disappear. It intensifies. A board that deploys AI without a structured decision record has no way to demonstrate oversight after an incident. With one, the board can prove it governed responsibly. The question is not whether accountability creates value. The question is whether the board can afford to operate without it.

The same liability they face for any failure of oversight. Corporate law does not distinguish between harm caused by a human employee and harm caused by an autonomous system acting on the corporation's behalf. Directors in the UK face personal exposure under the Companies Act, section 174. In Canada, the standard is the CBCA, section 122. The EU AI Act adds regulatory penalties on top of existing civil liability for high-risk systems. In every case, the court asks one question: did the directors take reasonable care? A tamper-evident decision record is how you answer yes.

No. That assumption is the single most expensive misconception in enterprise AI. The AgDR standard seals each decision in 620 nanoseconds. That is faster than a network packet leaves the machine. Accountability at this speed does not compete with performance. It removes the governance objection that has kept institutional capital on the sidelines. Organizations that can prove their AI is auditable deploy faster, insure more cheaply, and earn regulatory approval with less friction. Accountability is not the brake. It is the bridge between prototype and production.

They already do. No new legislation is required. Evidence law in most jurisdictions recognizes contemporaneous business records as presumptively reliable. Corporate law imposes a duty of care on directors regardless of whether decisions are made by humans or machines. The EU AI Act mandates traceability and logging for high-risk systems under Article 12. The NIST AI Risk Management Framework identifies transparency and accountability as core properties. These frameworks already define what is expected. What has been missing is the technical infrastructure to satisfy them at machine speed. That is what AgDR provides.

No. The standard was built and stress-tested in Canada, but the principles are universal. AgDR maps to the EU AI Act, the NIST AI Risk Management Framework, ISO/IEC 42001, and common-law fiduciary duty as practised across the UK, Australia, and other Commonwealth jurisdictions. Contemporaneous capture, tamper-evidence, chain of custody, and named human accountability are not Canadian concepts. They are evidentiary fundamentals recognized by courts everywhere. Canada is where the standard was proven. It is not the boundary of where it works.

Timing and integrity. A standard audit log is written after the decision, sometimes minutes or hours later. It can be modified without detection. An AI decision record is captured at the exact inference instant, cryptographically signed, and chained into a structure where any alteration is mathematically visible. Courts can rely on the second kind without expert reconstruction. The first kind leaves room for doubt. For regulators and fiduciaries, the distinction is simple: one is a note about what happened. The other is proof.

Directly and immediately. The EU AI Act requires providers and deployers of high-risk AI systems to maintain logging capabilities, ensure traceability, and provide human oversight. Article 12 mandates that systems produce records sufficient to reconstruct and audit each decision. Compliance obligations begin in August 2026. Organizations that lack structured decision records will face enforcement action, fines, and exclusion from the EU market. AgDR satisfies these requirements as a technical capability built into the inference pipeline, not as a policy document filed after deployment.

That is exactly what it is for. When an AI decision causes harm, the first question from regulators, insurers, and courts is the same: show us the record. A complete AgDR trail proves three things. It proves the AI was operating within defined boundaries. It proves a named human was accountable for oversight. And it proves the record existed before the incident, not after. This is the difference between explaining what you think happened and demonstrating what you know happened. Corporations that can show the record negotiate from strength. Those that cannot are left to explain why they chose not to keep one.

Completely open. The standard is dual-licensed under CC0 1.0 Universal and Apache License 2.0. You choose whichever licence suits your deployment. The specification, the SDK, and the reference implementation are all published on GitHub with no proprietary dependencies, no vendor lock-in, and no royalties. Accountability infrastructure should not be owned by any single company. It should be a public good, the same way evidence law itself is a public good. That is why the standard is open.

Every unrecorded AI decision is uninsurable liability. That is not a projection. It is the current state of the insurance market for autonomous systems. When something goes wrong, regulators will ask for the decision record. Courts will ask for the chain of custody. Insurers will ask whether the risk was auditable before the loss. If the answer to any of those questions is no, the corporation absorbs the full cost alone. Every profession that earned institutional trust did it by keeping the record. Medicine keeps charts. Aviation keeps black boxes. Law keeps transcripts. The only question is whether AI will join that list voluntarily, or be forced onto it after the first landmark ruling.