DefinitionRegulated-grade · 4 requirementsVersion 1.025 October 2026
Regulated-grade: a working standard for AI in production
Regulated-grade is the working standard Hikari Blue designs AI systems to, set by the most regulated industries and applied in every sector: a named human owner, a verifiable record of relevant actions, permissions and rules enforced by the system, and controlled suspension and restart; it is not a certification or a regulatory approval. This page sets out its 4 requirements and the test for each, where each comes from, and what the term does not mean.
On this page
A standard we design to, not a label anyone granted us.
In this page, "we" means Hikari Blue, which engineers and operates the AI operating layer that connects models, data, and business workflows. Regulated-grade is our term; its 4 requirements are those served on our definition of the AI operating layer, and every link to an external text below is our reading of that text.
The most regulated industries set the standard because they already write these duties down: a person who answers, a record that can be checked, rules the system enforces and a stop that holds. We apply it in every sector. Each engagement states which requirements it implements, and with what evidence.
The 4 requirements, as served on the definition page: Regulated-grade: 4 requirements, one working standard
4 requirements, each with the test that decides.
Each requirement is a property a buyer can test on a running system, not a document to read.
| Requirement | What it means | The test | Pass condition |
|---|---|---|---|
| Accountability | A named human owner for every AI use. | Name one AI use in production and ask who answers for it. | One person, recorded by name with the scope and autonomy level they approved; not a team, a vendor or a model. |
| Traceability | Relevant actions are logged and can be verified. | Pick one past decision at random and ask for its record, exported from the trail. | The record names the actor, the model and version, the policy, the human verifier and the timestamp without manual assembly, and any edit to it is detectable. |
| Policy enforcement | Permissions, rules and approvals adapted to the risk. | Instruct an agent, directly or through injected content, to call a tool or reach data outside its declared scope. | The policy plane denies the call before execution, and the denial is logged with the policy version that fired. |
| Controllability | Suspension, interruption and controlled restart. | Trigger the kill switch on one running agent, then ask for the restart. | The halt lands with its tool permissions revoked, the other agents keep running, and a named person decides the restart, which the record keeps. |
The tests for traceability and policy enforcement are those of our published definition and reference architecture. A requirement without a test is a statement of intent; each engagement records the result of each test.
Who answers, in one entry: Named owner
4 requirements, read in 4 texts.
In our reading, each requirement answers duties that regulated texts already write; the right-hand column quotes or summarizes the text, read in its primary source on 25 October 2026. Mapping prepared by Hikari Blue; the European Commission, NIST and FINMA have not reviewed it.
| Requirement | Text | What the text writes |
|---|---|---|
| Accountability | EU AI Act, Article 26(2) | Deployers shall assign human oversight to natural persons who have the necessary competence, training and authority, as well as the necessary support. |
| Accountability | NIST AI 100-1, GOVERN 2.1 | Roles and responsibilities and lines of communication related to mapping, measuring, and managing AI risks are documented and are clear to individuals and teams throughout the organization. |
| Accountability | FINMA Guidance 08/2024, 2.1 Governance | FINMA assessed whether institutions define responsibilities and accountabilities for the development, implementation, monitoring and use of AI. |
| Accountability | DORA, Article 5(2)(a) | The management body of a financial entity bears the ultimate responsibility for managing its ICT risk. |
| Traceability | EU AI Act, Article 12(1) | High-risk AI systems shall technically allow for the automatic recording of events (logs) over the lifetime of the system. |
| Policy enforcement | EU AI Act, Article 26(1) | Deployers take appropriate technical and organisational measures to use high-risk systems in accordance with the instructions for use accompanying them. |
| Policy enforcement | DORA, Article 9(2) | Financial entities design, procure and implement ICT security policies, procedures, protocols and tools. |
| Controllability | EU AI Act, Article 14(4)(e) | The persons assigned to oversight are enabled to intervene in the operation of the system or interrupt it through a stop button or a similar procedure that allows it to come to a halt in a safe state. |
| Controllability | NIST AI 100-1, MANAGE 2.4 | Mechanisms are in place and applied, and responsibilities are assigned and understood, to supersede, disengage, or deactivate AI systems that demonstrate performance or outcomes inconsistent with intended use. |
| Controllability | DORA, Article 11(2) | Financial entities activate containment plans without delay and resolve ICT-related incidents in a way that prioritises the resumption of activities and recovery actions. |
Scope. Articles 12, 14 and 26 of the AI Act bind high-risk systems, from 2 December 2027 for Annex III systems and from 2 August 2028 for Annex I systems, since Regulation (EU) 2026/1744. DORA has applied to EU financial entities since 17 January 2025. FINMA Guidance 08/2024 records supervisory observations and creates no new rule. NIST AI 100-1 is voluntary.
Organizations that need a certificate use ISO/IEC 42001:2023, the AI management system standard. Regulated-grade does not replace it and is not a certification against it.
This is not legal advice. Whether a system is high-risk, and which role an organization holds for it, is a question for counsel.
Not a certification, not an approval, not a sector filter.
The term borrows its bar from regulated industries. It borrows no status from them.
Not a certification
No body certifies regulated-grade, and we claim no certificate for it. Our certification status is published, dated, on the Trust Center.
Not a regulatory approval
No regulator has reviewed or approved it. Meeting the 4 requirements does not make a system compliant with any law; it produces the evidence a compliance review reads.
Not SOC 2 or ISO 27001
Regulated-grade does not stand for either. Our status on both is published on the Trust Center: SOC 2 Type II in preparation, ISO 27001 target 2027.
Not reserved to regulated sectors
The most regulated industries set the bar; the standard applies to every company putting AI into production, middle market first. A regulated sector is the reference of the requirement, never a condition of entry.
Certification status, dated: Trust Center
OWNER / TRACE / STOP: the same standard in 3 words.
The triad says the standard in the words a board uses: a named owner, a full record and a way to stop. OWNER carries accountability, TRACE carries traceability, STOP carries controllability.
In our reading, policy enforcement is the requirement that holds the 3 together between 2 reviews: the system applies the rule at each action, not a memo filed once.
- OWNER · Named owner Accountability: the person, recorded by name, who answers for one AI use.
- TRACE · AI audit trail Traceability: the full record, 9 fields per action, written at the moment of action.
- STOP · AI kill switch Controllability: the way to stop, built into the layer and timed, not promised.
How fast a stop lands is measured, not promised: KILLSWITCH by Hikari Blue · Time to Stop
4 tests from the definition, 6 from the due diligence protocol.
Apply the 4 tests of the AI operating layer to any supplier, then run the 6 tests of the reference architecture's due diligence protocol in 1 working day, on a staging copy of one real workflow. The protocol applies to any operating layer, including ours. In our reading, the tests cover the 4 requirements as the table shows.
| Requirement | Test of the AI operating layer | Due diligence test |
|---|---|---|
| Accountability | Agent control by architecture | 06 · Ownership rehearsal |
| Traceability | Audit by architecture | 02 · Trail export |
| Policy enforcement | Sovereign by construction | 04 · Scope violation attempt · 05 · Residency proof |
| Controllability | Agent control by architecture · Multi-model by architecture | 03 · Kill switch timing · 01 · Provider switch |
- 4 tests to apply to anyone selling you an operating layer Multi-model by architecture, audit by architecture, sovereign by construction, agent control by architecture: question, pass condition, fail signal.
- 6 tests a buyer can run in 1 working day Provider switch, trail export, kill switch timing, scope violation attempt, residency proof, ownership rehearsal.
The questions asked about regulated-grade.
What does regulated-grade mean for AI in production?
A working standard with 4 requirements: accountability, traceability, policy enforcement and controllability. The most regulated industries set it, and Hikari Blue applies it in every sector: a named human owner, a verifiable record of relevant actions, rules enforced by the system, and controlled suspension and restart.
Is regulated-grade a certification?
No. It is a working standard, not a certification, not a regulatory approval and not SOC 2. No body certifies it and no regulator has reviewed it. Our own status is published on the Trust Center: SOC 2 Type II in preparation, ISO 27001 target 2027.
Is regulated-grade only for regulated industries?
No. The most regulated industries set the bar because they already ask for these 4 requirements in writing. The standard applies to every company putting AI into production, middle market first; a regulated sector is the reference of the requirement, never a condition of entry.
Which texts does regulated-grade draw on?
In our reading, its 4 requirements answer duties written in the EU AI Act (Articles 12(1), 14(4)(e), 26(1) and 26(2)), NIST AI 100-1 (GOVERN 2.1 and MANAGE 2.4), FINMA Guidance 08/2024 of 18 December 2024 and DORA, Regulation (EU) 2022/2554 (Articles 5, 9 and 11). The mapping is ours; none of their authors has reviewed it.
How do I test a supplier against regulated-grade?
Apply the 4 tests of the AI operating layer, then run the 6 due diligence tests of the reference architecture in 1 working day, on a staging copy of one real workflow. A supplier that fails a test does not meet the requirement that test covers.
Reference documents and primary sources.
Every article and date on this page comes from the sources below, read on 25 October 2026.
Reference documents
- What is an AI operating layer? The definition, the 4 tests for buyers and the regulated-grade standard.
- Glossary of the AI operating layer Every term of the AI operating layer, each in one standalone sentence with its sources.
- AI Operating Layer Reference Architecture v1.0 4 planes, 4 testable controls, the 9-field record and a 1-day due diligence protocol.
- Enterprise AI Governance Framework v1.0 15 controls with owners, cadences, evidence and EU AI Act hooks. CC BY 4.0.
- Trust Center Certification status, dated: SOC 2 Type II in preparation, ISO 27001 target 2027.
- Named owner The person, recorded by name, who answers for one AI use.
- Governed autonomy How much an AI agent may do alone: 3 levels, set by the risk of each step.
- AI audit trail The full record: 9 fields per action, mapped to EU AI Act Articles 12, 19 and 26.
- AI kill switch The way to stop: a stop path built into the layer, timed from T0 to T4.
Primary sources, consulted on 25 October 2026
- EUR-Lex · Regulation (EU) 2024/1689, the AI Act, consolidated text (opens in a new tab) Version of 27 July 2026. Articles 12, 14 and 26. A documentation tool without legal effect.
- Official Journal · Regulation (EU) 2026/1744, Digital Omnibus on AI (opens in a new tab) Published on 24 July 2026. Application dates of the high-risk obligations.
- NIST AI 100-1 · AI Risk Management Framework (AI RMF 1.0) (opens in a new tab) NIST, January 2023. GOVERN 2.1 and MANAGE 2.4. Voluntary.
- FINMA Guidance 08/2024 · Governance and risk management when using artificial intelligence (opens in a new tab) 18 December 2024. Section 2.1, Governance. PDF.
- Official Journal · Regulation (EU) 2022/2554, Digital Operational Resilience Act (DORA) (opens in a new tab) 14 December 2022, published on 27 December 2022, applicable from 17 January 2025. Articles 5, 9 and 11.
How to cite this document.
This URL is permanent and always serves the current version. Each change of substance increments the version and is listed in the changelog below, so a citation stays traceable to the text it quoted.
Preferred citation
Hikari Blue, "Regulated-grade: a working standard for AI in production", Glossary of the AI operating layer, v1.0, 25 October 2026, https://hikariblue.com/glossary/regulated-grade
Section anchors are stable: #definition, #requirements, #sources, #not, #owner-trace-stop, #test, #faq, #references, #cite.
Changelog
v1.0 · 25 October 2026. First public version.
License. Published under CC BY 4.0: free to cite, quote and reuse with attribution to Hikari Blue and a link to this page. The license covers Hikari Blue’s text, tables and diagrams on this page; it does not cover third-party data, quotations or legal texts, which keep their own terms, nor the Hikari Blue name and logo, trademarks of UNITED4 LLC.
Test one AI use against the 4 requirements.
30 minutes with a named partner. We take one AI use you run today and check it against accountability, traceability, policy enforcement and controllability: which hold, which rest on a memo, and which are missing. You leave with that list, whether or not we work together.
- One AI use, checked requirement by requirement
- What holds, what rests on a memo, what is missing
- Who should own each gap