A determination of whether a competent human was positioned, informed, authorized, and capable of materially changing a consequential AI-mediated decision before consequence occurred.
When an AI system shapes a decision that costs someone money, coverage, credit, or liberty, and that decision is later challenged, the organization is asked one question. Did a qualified human exercise meaningful judgment before the outcome took effect, and can you prove it?
Almost no organization can answer with evidence rather than assertion. A system log records that a person clicked approve. It does not record whether that person had the information to disagree, the authority to override, or the time to do either.
An approval is not oversight. The two are indistinguishable in every audit trail in production today, and they stay indistinguishable right up to the moment a court, a regulator, or a carrier asks for the difference.
Three conditions hold across every AI-mediated decision environment examined to date.
System logs show that a person approved something. Not that they had the information, the authority, or the time to meaningfully change it.
The two produce identical records. The difference surfaces only in litigation, an audit, or a claim dispute, at which point it cannot be remedied retroactively.
Generic AI governance frameworks assess the AI system. Almost nothing measures whether the human control around one specific decision was real.
Conclusion. You can't blame the machine, and you can't fake the human. Both halves of that sentence are now legally operative.
HOP is grounded externally. Every criterion traces to binding law, regulatory guidance, audit-evidence doctrine, or published human-factors research. None of it is an internally invented rubric.
| Citation | Authority | Operative for |
|---|---|---|
| Civ. Code § 1714.46 | California AB 316Ch. 672, Stats. 2025. Signed 13 Oct 2025. | Removes the defense that an AI system acted autonomously. Liability reaches any party that developed, modified, or used it. In force. |
| C-634/21 | OQ v Land Hessen (SCHUFA)CJEU, ECLI:EU:C:2023:957. 7 Dec 2023. | Nominal human involvement does not remove a decision from meaningful-oversight requirements. Involvement must be functional. |
| Reg. (EU) 2024/1689 | EU AI Act, Arts. 14 & 26As amended, post Digital Omnibus. | Effective human oversight and competent, authorized overseers. Annex III from 2 Dec 2027; Annex I from 2 Aug 2028. |
| AS 1105 | PCAOB, Release 2024-007Adopted 12 Jun 2024. | System-generated information does not self-authenticate and must be independently evaluated. The doctrinal spine of evidence discipline. |
| TechDispatch 2/2025 | European Data Protection Supervisor23 Sep 2025. | Criteria for effective human oversight of automated decision-making. |
| ISA 500 / AU-C 500 | IAASB & AICPAAudit evidence. | Sufficiency and appropriateness of evidence; the standard HOP borrows its evidence caps from. |
Regulation strengthens the case. It isn't the case. AB 316 and SCHUFA are in force now.
HOP evaluates four conditions in sequence. Each is necessary. None is sufficient alone.
Was the human placed at a point in the workflow where intervention was still possible before consequence occurred?
Did they receive what they needed to disagree, including the basis for the AI's output and its known failure modes?
Did they hold actual authority to override, and was overriding a survivable act inside the organization?
Were they competent in the matter, and was the workload compatible with judgment rather than throughput?
Dimensions, thresholds, and gate logic are controlled methodology and are not published. We disclose the construct we test, never the scoring mechanics. Publishing the mechanics would let a client engineer the result, which is the failure mode every self-assessment framework already has.
PromptID never implements the controls it assesses. Not as a policy preference, as a structural bar. Finance separates auditors from the books. Pharma separates the FDA from the manufacturer. Most advisory firms cannot make this claim because implementation is where their revenue is.
A HOP Determination is never a maturity score, a colour, or a percentage. It resolves to one of three findings.
The human was positioned, informed, authorized, and capable. The evidence proves it.
The oversight conditions were not met. The record shows what was missing and where it broke.
Oversight may have occurred. Nothing survives that could prove it. In an audit this is the same as no.
An instrument that cannot return a bad result is not an instrument.
A determination is only worth what its record is worth. This page seals itself. The digest below is a real SHA-256 computed in your browser over the full text of this document, right now.
Change one character on this page and the digest changes completely. That is the whole idea behind a frozen Decision Record: not that the evidence is secret, but that it cannot be edited after the fact without the edit being obvious.
Who authorized this?
One day someone will ask that about a decision your organization already made. The answer is either an evidence package or a shrug.