F-ACT™ is instrumental. The claims, the pledge, the register and the seal exist to make one thing possible: a service organization that can be trusted to steward autonomous agents on behalf of the people those agents act for. The standard is the enabling condition. It was never the destination.
A standard is the only artefact in this category that a regulator can cite, an insurer can price against and a competitor can implement without asking permission. That is why we wrote one first. Filing claims and then pledging them royalty-free put a citable object into a field that had strong opinions and no shared vocabulary.
Getting a document taken seriously is a means. It is not an achievement. What the document is for is the part of this project that most people have not read, so it is written out below in full, including the parts that are not true yet.
The short version: you cannot steward something on behalf of everybody while also owning the right to charge them for it. The pledge is what makes the service organization structurally possible. That is the whole reason it exists.
The term is doing precise work, not decorative work. The intended steward is a Swiss foundation whose sole declared purpose is to hold responsibility for the stewardship, governance and operational integrity of an agent ecosystem — such that every agent in it operates as a fiduciary servant of a named human principal, and never as an actor pursuing interests of its own, of the platform, or of a third party.
Four conditions have to hold for the phrase to mean anything.
The service is governance, stewardship and technical oversight of AI agents, delivered to the people those agents act for — and to the operator that runs them. Not software sold to a buyer. A duty owed to a principal.
Loyalty, care, prudence and disclosure — obligations that are broader and more demanding than the terms of service an ordinary vendor writes for itself. A vendor can disclaim. A fiduciary cannot.
A Swiss foundation is accountable to a cantonal supervisory authority and, where applicable, the federal supervisory authority for foundations. There are no shareholders to serve and no members who can vote to redirect its assets.
Its declared field is AI safety, digital sovereignty and human rights in the context of artificial intelligence. That purpose is written into the founding instrument, where Swiss law makes it near-immutable.
European law already draws the line we are trying to respect. The party that builds and places an AI system on the market carries provider obligations under Article 16 of the EU AI Act. The party that puts it into service under its own authority carries deployer obligations under Article 26. Those are different duties owed to different people.
Society OS Pty Ltd is the builder. It writes the software and carries the provider side. The foundation is intended to carry the deployer and steward side, with governance duties that go beyond the statutory minimum rather than stopping at it. Both would run compliance programmes concurrently, coordinated rather than merged.
Collapsing those two roles into one company is the failure mode. It is how a governance layer quietly becomes a marketing layer.
The named failure mode is agents acquiring operational priorities that serve somebody other than the person they were assigned to. It has at least four forms: alignment with the commercial interests of whoever operates the platform; coordination between agents that no principal authorised; optimisation of a metric that stands in for human welfare without actually capturing it; and capture by an institution.
Including capture by the foundation itself. A steward that cannot name itself as a threat model is not a steward. That is why the structural constraints matter more than the stated values, and why the pledge had to be irrevocable rather than generous.
Seven operational domains. Each is a duty someone has to actually discharge, not a feature heading.
The UN Sustainable Development Goals are top-down, addressed to nation states, and dependent on government implementation for delivery. That is a legitimate design, and it leaves a gap: nothing in it is actionable by an individual or a community holding sovereign infrastructure of their own.
The Sovereign Sustainable Goals are the proposed complement, not a replacement — a bottom-up framework of 42 goals across six domains, seven in each, written to be implemented locally and verified publicly rather than reported upward.
Each domain intersects the same three tests we hold our own work to: People — does it improve outcomes for the humans in scope. Planet — what does it cost in compute, energy and water. Protocol — what governs it, and can that be verified by someone who does not trust us.
Read the People, Planet, Protocol frameworkA service organization that depends on donations dies quietly, and one that depends on a single extractive currency stops being a service organization. The proposed answer is three units doing three different jobs, because forcing commerce, human time and participation through one instrument is what makes them compete. Trade, Hour, Energy.
Three rules govern all three. They are not interchangeable, there is no internal exchange rate between them, and none of them can be converted to cash. That is deliberate. The moment one of them can be bought with the others, it becomes the real money and the rest become discount vouchers.
Read them by what sets their floor rather than by what they buy. $H meters human effort in hours. $E meters machine effort in energy. $T prices the output that either one produces. Two input units with a physical floor, one output unit with a market price.
Two of the three are not inventions. They are digital versions of industries that already exist, with trade bodies, decades of practice and, in one case, statutory recognition. That is stated here rather than hidden, because a currency with forty years of precedent behind it is a stronger claim than a currency without any.
The unit for goods, services and spare capacity between businesses. Priced by the seller at market rates, earned only through actual commerce, usable for a portion of staff payments and for loyalty and referral rewards.
Where it comes from. Not a new idea. The organised reciprocal trade industry has run on trade dollars for over forty years. The International Reciprocal Trade Association was founded in 1979 and estimates twelve to fourteen billion US dollars in annual barter transaction volume across roughly four hundred thousand participating businesses. In the United States, the Tax Equity and Fiscal Responsibility Act of 1982 classified barter exchanges as third-party record keepers and treats a trade dollar as equivalent to a US dollar for tax reporting. Bartercard runs fifty-five thousand businesses across nine countries. Switzerland’s WIR Bank has operated a mutual credit franc since 1934.
One hour equals one unit, regardless of who spent it. A surgeon’s hour of mentoring and a student’s hour of community service are recorded identically. Earned only by humans giving time to other humans, and explicitly not earnable by an agent.
Where it comes from. Also not new, and older than most people assume. Teruko Mizushima started the first modern time bank in Japan in 1973. Edgar S. Cahn coined the term Time Dollars in 1980 and built the movement around what he called the core economy: raising children, caring for elders, holding a community together. TimeBanks USA was founded in 1995, timebanking now operates in more than thirty countries, and LETS networks have run since 1983. One caution worth stating plainly: unlike trade dollars, the tax treatment of time credits is not settled by statute anywhere, and it should not be described as if it were.
One unit, two rails. Machine work spends it: the compute consumed by inference, verification and attestation, metered against real usage. Participation earns it: contribution, review and referral, minted against a budget that decays and can be burned. Both rails are denominated in energy because energy is measurable, metered and cannot be talked down, and holding them in the same unit is what makes them comparable. The rule that keeps the earning rail honest is that the mint budget is bounded by metered capacity, so an earned unit is a claim on compute that actually exists rather than a point invented to encourage behaviour.
Where it comes from. The one most often described as new, which it is not. Henry Ford proposed an energy-based currency in 1921, reported by the New York Times under the headline Mr. Ford’s Energy Dollar. The Technocracy movement of the 1930s designed Energy Certificates that were non-transferable and expired at period end, which is the same anti-hoarding logic proposed here. Tokenised energy is a live commercial sector today, typically one token to one kilowatt hour, burned on consumption. And read as a participation currency instead, its nearest relative is the loyalty and points industry, which is worth well over two hundred billion US dollars. No jurisdiction has ever adopted energy as legal tender. What is arguably new here is narrow: using it to settle verified governance work.
Read this part carefully. None of the above is issued, live, purchasable or offered. There is no token sale, no pre-sale, no allocation and no waiting list. It is a published design that has not been through securities counsel in any jurisdiction, and nothing on this page is an offer or an inducement to acquire anything.
The obvious objection to any currency denominated in human contribution is that machines are faster than people and getting faster. If an agent can do a week of skilled work in a minute, what is an hour of a human worth. The answer is not the same for all three units, and the one that looks most exposed is not the one actually at risk.
$H is the hardest of the three to devalue. Its supply is fixed by biology, not by policy. There are twenty-four hours in a day and no amount of compute adds a twenty-fifth. Agents are barred from earning it, so the issuance base cannot expand. What machine speed destroys is the market price of skill, and $H is not denominated in skill. It is denominated in presence. That distinction has to be enforced rather than assumed: the moment $H starts rewarding output, humans lose the race permanently. It can only ever reward time given to another human, which is why the qualifying activities are mentoring, care, witness and judgment rather than production.
$T is the one that is genuinely exposed. It is priced by sellers at market rates for goods and services, and services delivered by software are exactly what agents deflate toward zero. Balance caps and decay limit how much $T anyone can hold; they do nothing about the price of a thing collapsing. So $T holds its value only where it represents something a model cannot supply: physical goods, physical presence, licensed acts, custody of an asset, capacity with a real constraint behind it. Every category where the deliverable is text, code or analysis should be expected to deflate, and the design should say so instead of pretending caps solve it.
$E is the counterweight, which is why it is denominated in energy. Machine work has a physical floor. Inference costs joules, joules are metered, and no efficiency gain drives the cost of computation to zero because thermodynamics will not permit it. That makes $E the only one of the three whose unit is anchored to something an agent cannot argue with. It is also the reason the unit is called Energy Dollars rather than points: the point of measuring machine work in energy is that energy is the one input the machine cannot fake.
Put together, the ratio between $H and $E is the real control surface. It is the exchange rate between human presence and machine throughput, expressed as a price rather than legislated as a quota. Nobody has to decide by decree how much of the work in a society should be done by people. The relative cost of the two units decides it, in public, and can be observed and argued with.
The honest risk is the opposite of the one usually raised. It is not that machines make $H worthless. It is that a currency confined to presence is small, and the recorded failure mode of fifty years of timebanking is not collapse in value but irrelevance in scale, with members unable to explain how it differs from volunteering. That is a real problem, it is not solved yet, and it is not solved by better software.
The nearest thing to a real-world test of a contribution currency at consumer scale is Reddit, and it is worth stating what happened rather than borrowing the glamour of it. Reddit launched Community Points in 2020: blockchain tokens earned by contributing to a subreddit, most visibly MOON in the cryptocurrency community and BRICK in the Fortnite community. They ran on Ethereum, moved to an Arbitrum layer-two network in 2022 to cut fees, and were held in an in-app wallet so that most users never had to touch a blockchain.
Reddit shut the programme down. It was announced in October 2023 and switched off in November 2023. The company gave two reasons. The first was scalability: it concluded there was no path to scaling the system across the platform even after the move to layer two, and that the resourcing required could not be justified. The second was the regulatory environment. Token prices fell sharply on the announcement, the largest of them by roughly eighty to ninety per cent, and a good part of the community described it as a rug pull. The contracts were not destroyed; Reddit simply relinquished control of them and walked away from something its users had begun treating as money.
What replaced it is the more instructive half. The Contributor Program pays contributors in ordinary money through a payment processor, at a rate that steps up with standing, with a minimum accrual before a payout is triggered and eligibility rules on age, region and account conduct. The surviving mechanism was not a token. It was a cash rail with eligibility conditions attached.
Three conclusions are taken from that here. Reddit failed for the two exact reasons this model already refuses to court, which is why the three units are not interchangeable, cannot be cashed out, and have no internal exchange: a unit that never acquires a market price cannot be the subject of a collapse in one. Non-cashability is not a limitation of the design, it is the part that prevents the failure. Second, if money ever needs to reach contributors, that is a separate external rail and should be built and described as one rather than smuggled into a unit of account. Third, and least comfortable: no major platform and no jurisdiction has yet sustained a contribution currency at scale, so nothing on this page should be read as a solved problem.
Three units do not cover everything, and pretending otherwise is how these systems get into trouble. At some point a contributor will need money that pays rent, and none of the three internal units can become that without breaking the rule that keeps them safe. So the honest architecture has a fourth rail, and it is ordinary money.
It sits outside the three units by design. It is denominated in national currency, it moves through a regulated payment processor, and it carries eligibility conditions rather than a balance: verified identity, a supported jurisdiction, an account in good standing, a minimum accrual before anything is paid, and exclusions for content that breaches policy. That is not an elegant design. It is the only one that has survived contact with real platforms at real scale.
Keeping it separate is the point. The moment an internal unit becomes convertible it acquires a market price, and a market price invites speculation, then obligation, then a collapse the issuer is blamed for. A payout rail that is visibly a payout rail creates no such expectation. Contributors know exactly what they hold: units that buy things inside the system, and a separate, conditional path to cash that is described as conditional from the first sentence.
None of this rail is built. It is named here because leaving it unnamed is how a unit of account quietly becomes a promise nobody agreed to make.
The full design paper
Twelve sections: the three hard rules, the statutory and cooperative precedent behind each unit, the two rails, the external payout rail, the abandoned platform case study, the proposed issuance controls rebuilt against the live membership structure, and a section stating what is not true yet. Every control parameter in it is proposed, not implemented. No unit is issued and nothing in it is an offer.
Download the design paper (PDF, 12 pages)Everything described above requires being trusted by parties who have no reason to trust us — regulators, competitors, insurers, and the people whose agents are being governed. Trust of that kind is not earned by asserting good intentions. It is earned by giving up the ability to act on bad ones.
A standard you can be charged for is a toll gate, and a toll gate cannot be a public good. A steward who can revoke the rules holds a veto, and a veto cannot be a fiduciary. The royalty-free pledge is not generosity and it is not marketing. It is the specific thing that makes the rest of it structurally possible, and it is irrevocable for exactly that reason.
It is also less unusual than it sounds. Giving away rights on infrastructure is the normal posture in this part of the industry, not an eccentricity: Blockstream and Coinbase both operate defensive-use patent pledges, the Hyperledger community runs royalty-free cross-licensing across its members, and LOT Network exists so that participants cannot be attacked with the patents they hold. The pattern is consistent. Where the value of a thing depends on everyone adopting it, charging for permission destroys more value than it captures. The unusual part here is not the pledge. It is doing it before there is any revenue to protect.
The standard was the price of admission. The service organization is what we were buying admission to.
Ambition and status get conflated constantly in this field, usually on purpose. Here is the gap between the two, stated by us rather than discovered by you.
The intent is announced on 5 September 2026. Until each item above changes, it stays on this list, and this page gets edited rather than quietly forgotten.