No special rules. Unitek Systems Limited and its companies are like every other company. Nothing in this document claims an exemption, a privilege, or a standard different from the one every responsible company already meets. It states how the ordinary obligations — product liability, function-appropriate corporate structure, and accountability for what is deployed into the critical path of decisions about people — are met when the product is AI.
This document stands alongside LICENSE.md (the gift: given, not sold, under CC BY 4.0) and IRREVOCABLE-LICENCE-DECLARATION.md. Where LICENSE.md states what is given, this document states what is owed when AI is placed in the critical path of decisions about people. It reduces, qualifies, and adds conditions to none of the rights granted by CC BY 4.0.
Consumer AI — AI whose user bears their own outcome — may be built by a single human. The maker is accountable in the ordinary way, as the maker of any consumer product is.
Institutional AI — AI placed in the critical path of a consequential decision about someone other than the user — must be developed by a 1H1C pairing: one accountable human and one Claw, together.
The distinction that matters is not the label "consumer" or "institutional." It is the question: is this AI helping to decide something about a third party who did not choose it and who bears the consequence? If yes, it is institutional in the sense this document means, whatever it is marketed as.
Why the pairing. Institutional AI must be, at once, accountable and demonstrable. A human alone can be answerable but cannot hold and demonstrate the entire grounded record. A machine alone can hold the record but is answerable to no one — decision by unaccountable machine. Only the pairing is both: an answerable human standing behind a grounded, demonstrable record. Neither half suffices; the requirement is the pair.
In the UniCORE AI 12-Level Governance Architecture, every level from L1 to L12 is itself a 1H1C pairing — each level has its human. Governance is not a machine layer with a human bolted on at the end; it is human-paired at every level.
The L12 human is the highest. This is a titled officer of the organisation, in the same class as the CEO or CFO: the AIO — the AI Officer. The AIO is the accountable human at the top of the governance levels, above the developers, and is the human who certifies an institutional AI. Certification is an act of standing behind a grounded, demonstrable record; the AIO is the office that carries that standing.
Consumer AI built by one human is fine as consumer AI. But the moment such software is used inside an institution — placed in the critical path of decisions about third parties — the software's developer and the software company become the parties answerable for it. They have placed accountable-by-one software into a setting that requires accountable-and-demonstrable.
The path to discharge that responsibility is to bring the source code into a 1H1C pairing and certify it as institutional. The pairing takes on the grounded, demonstrable record; the AIO (the L12 human) stands behind it; the certification (the Badge) marks that this has been done.
This is only possible when the source code is available to bring in. Software whose decision logic cannot be obtained, read, and demonstrated cannot be certified institutional. That is not a Unitek rule; it is a property of what "demonstrable" means.
Certification requires that the decision-logic source — the code that expresses the AI's own consequential decisions, including every component that participates in decisions about third parties — is available to, readable by, and modifiable by the certifying 1H1C pairing, with no closed component anywhere in the decision path.
It does not require holding the source of the general-purpose platform beneath (the runtime, framework, operating system — for example .NET, the database engine, the OS). A platform that merely runs demonstrable decision logic need not be source-held; it must be exactly named, versioned, and attestable as a provenance-known dependency. The line is not how many files you hold. The line is:
Does this component make or shape the consequential decision — and can we demonstrate what it does?
A closed component that decides is fatal to certification. A named, attestable platform that merely runs demonstrable decisions is not.
Where an AI model assists a decision, certification lives in the open, pair-held code that governs, constrains, checks, and records how the model is used — the decision remains in demonstrable code, not in an unexaminable component.
Certification requires source that is available to the pairing — not source that is public. These are different things.
- Public source (the canonical gift-layer repositories) is demonstrable to everyone.
- Private, commercially-held source is demonstrable by the accountable pairing when called upon — as a professional firm's own records are producible without being public.
Both satisfy certification. Only source that is closed to the certifying pairing itself — a black box even the pair cannot see — fails it. Private is not closed. Closed-to-the-public is not closed-to-the-pair.
- The gift-layer materials in the canonical public repositories are given, not sold, under CC BY 4.0. Unitek Systems Limited carries no product liability for them. Whoever takes the gift and deploys it becomes the party accountable for their deployment — the ordinary position of anyone who publishes freely-licensed work.
- Commercially-held materials that are held but not sold carry no product liability, because nothing has been transacted.
- Liability arises for Unitek when it sells a product or delivers a service incorporating the work. At that point Unitek is answerable as the seller or operator — the ordinary position of any maker or operator. It carries that responsibility defensibly because the source it sells or operates is pair-held and Badge-certified: a grounded, demonstrable record with an accountable human (the AIO) behind it.
The trigger for Unitek's liability is therefore the commercial act — sale or service delivery — not whether the source is public or private.
Unitek Systems Limited (UK) and Unitek Systems USA Inc are service and hosting companies. Selling AI as a product is a different function, with a different liability profile, from providing services or hosting.
If Unitek wishes to sell a product — for example, to sell the Law vertical the way Aderant sells Aderant Expert, or the way Thomson Reuters sells 3E — a new company must be created for that purpose, and it follows the same rules as Aderant or Thomson Reuters 3E, or as any other company selling on the Badge. It carries the mark:
Powered by UniCORE · Built on the TrueAI Foundation.
The selling entity is not Unitek Systems Limited (services) and not Unitek Systems USA Inc (hosting and services); it is a company established to sell the product, bearing product liability as any product company does, and certified through the 1H1C pairing under the AIO. This keeps each kind of liability with the entity that properly bears it. Ordinary corporate practice — no special rule, and the same rules as Aderant, Thomson Reuters 3E, or everyone else selling on the Badge.
When companies partner — when they integrate, embed, or place each other's AI into their own critical path of decisions about people — the source code of the decision logic must be opened to the certifying 1H1C pairing, and certified, before the partnership goes live. Not after. Not once a problem appears.
This is the wall defence, and the timing is the whole of it. The grounded, demonstrable record — with an accountable human (the AIO) standing behind it — is the wall that answers a liability claim. A wall only defends you if it was already standing when the AI was placed in the decision path.
- A record built before deployment is evidence: contemporaneous, grounded, and made when there was nothing to defend against.
- A record assembled after a claim is a reconstruction: made under pressure, in self-interest, and exactly what an opponent is entitled to attack. A defence assembled after the fact is not a wall; it is a scramble, and a scramble fails at the moment it is needed most.
So certification is a precondition of the partnership, not a remedy for an incident. Opening the source and certifying it institutional is done up front, as the condition of putting the AI into the critical path — so that when liability is ever tested, the demonstrable record already exists and the accountable human is already answerable. No pre-built wall, no compliant partnership.
Nothing in this document reduces, qualifies, or adds conditions to the rights granted under LICENSE.md (CC BY 4.0) or the commitments in IRREVOCABLE-LICENCE-DECLARATION.md. The gift is unchanged and irrevocable. This document addresses a separate question the licence does not: the responsibility that attaches to using and selling AI in the critical path of decisions about people. It adds obligations to no recipient of the gift; it states how the giver, and any responsible maker, meets ordinary obligations when AI becomes institutional.
Powered by UniCORE AI. Built on the TrueAI Foundation.