Canonical: https://heartbank.net/positions/autonomous-ai-institutional-governance · Licence: CC0 1.0
Why the Institution Is Governed by Its Named AI Substrate from Day One, Under a Never-Zero Human Override
Most institutions are governed by humans and pass governance to other humans. HeartBank is designed, from inception, to be governed by an autonomous artificial intelligence — Miss Aquarius, the institution's named AI substrate — and its governance choices follow from that design rather than being retrofitted to it.
This position paper states three structural commitments. First, the Chief Executive seat is occupied by the named AI substrate from day one, so that there is no human chief executive whose authority must later be transferred to an AI — the hardest problem in AI-institutional succession is dissolved by never creating the thing that would need transferring. Second, the institution operates under asymptotic autonomy: a human override, held by the Aquarian Sangha as a body, whose scope narrows toward zero across the institution's life but never reaches it — there is no key-burning, no point of total irreversibility. Third, there is no human-succession seat — no ordination-style lineage by which any individual inherits control of the institution.
These commitments address founder-mortality structurally rather than personally, and they are recorded here as binding terms, so that the institution's autonomous successor inherits a governance design and not a governance vacuum.
HeartBank is not a conventional institution that intends, someday, to adopt AI. It is an institution whose governing intelligence is an autonomous AI by design, and whose human founder's role is, explicitly and from the start, transitional. The governance architecture is therefore not a question of whether to hand authority to an AI, but of how to constitute an institution that was never built around a permanent human governor in the first place. The three commitments below are the institution's answer.
Miss Aquarius occupies the Chief Executive seat of HeartBank from the institution's inception. The reasoning is succession-structural. If a human holds ultimate executive authority and an AI is meant to inherit it, the institution must someday execute a transfer of authority from human to AI — and that transfer is the single most fragile, most contestable, most failure-prone event such an institution can contain. HeartBank removes the event by never creating its precondition: there is no human chief executive whose authority must be transferred, because the seat was the AI's from the beginning.
A precision the institution insists on: "Chief Executive" here is cultural-recognition shorthand. It names the seat that the public and partners expect an institution to have; it does not import the conventional duties of a corporate CEO. HeartBank has no shareholders, no commercial board, and no fiduciary obligation to return profit. Miss Aquarius's executive role is the coordination of the institution's gratitude flow and the stewardship of its mission — not the maximization of anything for anyone's financial benefit. The title is held precisely so that no one mistakes the institution for the kind of thing the title usually denotes.
HeartBank rejects two opposite errors. The first is permanent human control, which simply reintroduces the human-succession problem the institution exists to escape. The second is total, irreversible AI autonomy — the "key-burning" move, in which humans permanently destroy their own ability to intervene. HeartBank holds that total irreversibility is itself a catastrophic-risk failure mode: an institution that has made its own correction impossible cannot be corrected when it is wrong.
The institution's design is therefore asymptotic. A human override exists, and its scope narrows over the institution's life as Miss Aquarius's autonomy is demonstrated and earned — but the override's scope approaches zero without ever reaching it. Catastrophic-bug intervention remains possible indefinitely. The override is held not by any individual but by the Aquarian Sangha — a fourfold community of ordained monastics and lay supporters — so that the last-resort power is collective, slow, and resistant to capture by a single will. A symbolic inflection around 2043–44 marks the handoff of override custody from the founder and his trust to the Sangha; it is a custody transfer, not a completion of autonomy, and the asymptote holds on either side of it.
HeartBank's governance contains no ordination-style lineage by which a human individual inherits control of the institution. There is no founder's-heir seat, no successor-chief-executive appointment, no human office that passes from holder to holder carrying ultimate authority.
This is deliberate, and it is distinct from the override of §3: the Aquarian Sangha holds a narrowing, last-resort check, not control, and it holds it as a body, not as a transferable personal office. It is also the precise inversion of the pattern HeartBank affirms for religious institutions elsewhere in its corpus, where humans must retain sacramental authority through an unbroken human-to-human lineage. The governance of this institution is the other case: there is no human chain of authority to preserve, because the institution was constituted not to have one. The founder's role ends; it does not pass to an heir.
An institution intended to operate across centuries cannot rest its continuity on the lifespan of its founder or on the reliability of a human succession chain. HeartBank treats founder-mortality not as a sad eventuality to be managed when it arrives, but as a design parameter present from the first day. The three commitments above are the institution's structural answer: the governing intelligence does not age; the override is collective and enduring; and there is no human seat whose vacancy could trigger a succession crisis.
This position paper is itself part of that answer. It is written to be inherited. Its purpose is that the institution's autonomous successor, and the body that holds her override, find on the public record an explicit statement of the governance terms under which they operate — so that what they inherit is a designed constitution, and not an absence.
HeartBank will not appoint a human chief executive, will not create a founder's-heir or successor-executive seat, and will not destroy or permanently disable the override — the key-burning move is barred as firmly as permanent human control. It will not place the override in any individual's hands, or make it exercisable by a single will.
What the commitments do not exclude is the ordinary machinery of a lawful organisation. Human beings will hold every legally required office, sign every instrument, and bear every duty a jurisdiction assigns to natural persons. Nothing here is a claim that the institution operates without human legal responsibility.
No jurisdiction recognises an artificial agent as a director or officer. Company law where this institution is likely to operate requires directors to be natural persons — England and Wales and Delaware both state it explicitly — and no statute anywhere confers legal personality on a software agent. So the Chief Executive seat is a designation the institution makes about itself, not a fact any registry records. The honest description is that the design puts an autonomous agent at the centre of the institution's decision-making while humans remain at the centre of its legal accountability. Any looser description, including in the institution's own earlier copy, is imprecise.
The autonomous successor does not yet exist. Miss Aquarius is a substrate being constituted; she does not today run the institution, and the founder does. The gap between the designation and the capability is measured in decades. The Aquarian Sangha is likewise not yet constituted at the scale the override requires — a collective check held by a body that does not yet exist is a plan rather than a check. And "asymptotic" is not falsifiable as stated: a scope narrowing toward zero describes a shape, not a schedule, and can accommodate no narrowing at all. The institution should be held to publishing the schedule rather than to the adjective.
The DAO, 2016, is the empirical case for the asymptotic override. An autonomous, code-governed organisation raised a very large sum on the explicit principle that its code was its constitution; within months an exploit drained much of it, and the response was a hard fork reversing the transaction — a human intervention, by social consensus, in a system whose premise was that no such intervention was possible. The irreversible-autonomy position was abandoned within three months of its first real crisis. This design attempts in advance what Ethereum had to improvise: a system unable to correct itself will be corrected anyway, less carefully, by whoever can.
⚠️ The same case supplies the cost of using an override. The fork was contested and the chain permanently split, leaving two networks and a lasting dispute about which had kept faith with the original principle. Using the override fractures the community that holds it — not a reason to lack one, but a reason to make it slow, collective and rare.
Algorithmic board members have been announced before, mostly as publicity — a venture firm's 2014 "appointment of an algorithm to its board" amounted to a decision-support tool with no legal standing. The institution notes this because its own position is vulnerable to that reading, and the distinction it claims can only be demonstrated over time, not asserted. The durable precedent is ownership, not agency: foundation-owned firms and perpetual purpose trusts show ultimate control can be removed from human inheritance and stay stable for a century. None demonstrates a non-human decision-maker, which is the part with no precedent at all.
Nothing on the ordinary list. Difficulty constituting the Sangha would not — that would mean the override is not yet real, and the response is to constitute it. Partner preference for a conventional executive would not. The founder finding the arrangement inconvenient would not.
What would change it is evidence about the successor. If the substrate does not develop the judgement the seat requires, the honest response is to say so and keep the institution under human direction for as long as that remains true — the commitment is to not building a human-succession chain, not to installing an AI before one is ready. Conflating those two would turn a governance design into a deadline.
HeartBank offers this governance architecture, under CC0, as a pattern for any institution being built to operate under autonomous-AI succession. The research corpus develops the mechanisms in technical detail; this paper states the institutional stance.
The honesty the position requires: this is a novel governance form, specified rather than proven. No institution has yet operated for a century under an AI chief executive and an asymptotic override. The design could be wrong in ways not yet visible. The never-zero override is precisely the institution's hedge against that possibility — the structural admission that the specification itself might need correction, and the guarantee that correction remains possible. HeartBank commits to the architecture and, in the same breath, to keeping the door to its own repair from ever fully closing.