The Standard That Governs AI Agents Now Belongs to Everyone

Why Today

World Humanitarian Day exists to honour people who risk their lives to help others. It is a reminder that the systems humanity builds should serve humanity — not the other way around.

Today we publish F-ACT 1.0 — the Framework for Agent Conformance & Trust — the first open, royalty-free, universally implementable standard for governing autonomous AI agents. And we are issuing a Call for Founding Signatories to build a multi-stakeholder consortium that will govern it.

This is not a product launch. It is a public commitment: the rules that govern how AI agents act on behalf of people should themselves be governed by people — not by any single company, government, or bloc.

The Problem F-ACT Solves

Autonomous AI agents are deploying into economies, governments, and critical infrastructure faster than any governance framework can contain them. The frontier labs are racing to build them. Regulators are racing to regulate them. Nobody has published a standard that does the one thing governance must do: answer the hard questions before the agent acts, not after.

Every existing AI safety initiative focuses on model weights and training-time alignment — the inputs. None address the operational gap: what happens after deployment, in real time, at the moment of execution.

That gap is F-ACT's territory.

What F-ACT Is

This is not a product launch. It is a public commitment: the rules that govern how AI agents act on behalf of people should themselves be governed by people.

F-ACT is a standard — a set of published, versioned, freely citable rules. At its core are five deterministic questions that every agent must answer before it acts:

| # | Question | Domain | |---|----------|--------| | A | Who authorised this action? | Authority | | S | What boundaries constrain it? | Scope | | D | What data may it touch? | Data | | A | Is every decision recorded immutably? | Audit | | R | Can a human revoke it instantly? | Revocation |

These five questions — ASDAR — are the normative core. Four conformance tiers (L0 Unattested → L3 Provable) define graduated rigour. The defining principle: govern before execution — not after.

The full standard is published at [compliance.sovereignsos.com/standard](/standard) and freely downloadable as a PDF specification.

The Patent Pledge — Broader Than You'd Expect

Every patent claim essential to the standard is covered by an irrevocable, royalty-free, worldwide patent non-assertion pledge. But our pledge goes further than the industry norm: it covers any use of the core claims, not just conformance. You do not need to implement F-ACT to benefit from the pledge. You do not need to conform. You do not need our permission.

The distinction is simple: the standard says what must be true. Our operational patents protect how we do it best. You are free to use, build on, or require the core claims by any method, for any purpose, without a licence from us.

This is not generosity. It is architecture. A standard that people fear to touch is a standard that doesn't spread. We want F-ACT everywhere — including in the hands of our competitors.

The full pledge: [compliance.sovereignsos.com/standard/patent-pledge](/standard/patent-pledge)

A standard that people fear to touch is a standard that doesn't spread. We want S-ACT everywhere — including in the hands of our competitors.

From Open Standard to Shared Governance

Publishing an open standard is necessary but not sufficient. If one company controls the pen, the standard is only as trustworthy as that company's restraint. We don't think that's good enough — for us or for anyone.

Today we are issuing a Call for Founding Signatories to form a multi-stakeholder consortium that will govern F-ACT collectively. The consortium is modelled on the bodies that made the internet, the web, and open-source infrastructure possible: W3C, IETF, OASIS, the Linux Foundation.

How it works

Phase 1 (now): Society OS retains the editorial pen while the standard stabilises through its early versions. Founding Signatories participate in working groups, submit comments, and shape the roadmap. This is how every successful open standard has bootstrapped: one author, many reviewers, community governance once the foundation is stable.

Phase 2 (governance activates): Here is the part we thought hardest about. The obvious trigger — "hand over control once enough people sign" — is the wrong one. Headcount measures enthusiasm, not legitimacy. We could gather a hundred signatures in a fortnight and it would prove nothing about whether the right institutions are at the table. So governance does not transfer on signatory count. It transfers when both of these are true:

• Legitimacy quorum: signatories represent at least 4 of the 5 constituencies — including at least three regulators or safety institutes and at least three frontier labs or tech companies. The serious players must actually be in the room.
• Maturity floor: the standard has reached a stable v1.x and at least twelve months have passed since publication (so, not before 19 August 2027). You do not hand over an immature spec.

And a hard backstop so this can never become a stalling tactic: control transfers to the consortium no later than 19 August 2028, quorum or no quorum. We cannot keep the pen indefinitely by claiming the conditions were never met.

At activation: the governing board is elected by consortium members. No single member holds a veto. The editorial pen transfers from the originator to the consortium. Society OS retains one permanent board seat with no veto — a voice, not a chokehold.

Headcount measures enthusiasm, not legitimacy. Governance does not transfer on signatory count.

These triggers and the backstop are hard commitments. They are not discretionary. They are not deferrable. They will be written into the charter of the Swiss foundation that hosts the consortium.

Who should sign

We invite Founding Signatories from five constituencies:

1. AI Safety Institutes & Regulators — national or multilateral bodies with a mandate to govern AI deployment. 2. Frontier AI Labs & Tech Companies — builders of the systems this standard governs. 3. Enterprise Adopters — organisations deploying AI agents at scale who need a governance baseline. 4. Academic & Civil Society — researchers, ethicists, and advocates for digital rights and AI alignment. 5. Global South & Underrepresented Voices — ensuring the standard serves all of humanity, not only the blocs that build the models.

Founding Signatories are not committing to implement F-ACT. They are endorsing the principle that agent governance should be open, deterministic, and collectively governed — and joining the table where the standard's future is shaped.

There is no fee. There is no contract. A Founding Signatory endorses the principle; implementation and membership formalise later.

Sign at [compliance.sovereignsos.com/consortium](/consortium)

Why This Matters Now

The frontier AI labs, governments, and standards bodies are all racing to define the governance framework for autonomous agents. The Frontier Model Forum. The AI Safety Summits. The AI Safety Institutes. Everyone is building their version of the answer.

The race is not to build the biggest AI. It is to build the governance architecture that makes the biggest AI trustworthy.

And the clock is no longer abstract. EU AI Act enforcement for general-purpose AI began on 2 August 2026, with penalties reaching €15M or 3% of global turnover. ISO/IEC 42001 became a European standard this year. Identity protocols are being standardised — the Linux Foundation now stewards the Model Context Protocol; NIST has launched an AI Agent Standards Initiative. The obligations are real and multiplying. What is still missing is an open, citable standard for proving that a specific agent action was authorised before it happened. That is the gap F-ACT fills.

If the answer that wins is proprietary — controlled by a single company or a single bloc — the cure becomes another disease. The world does not need another walled garden dressed as governance.

F-ACT exists to be the technical standard that treaties, regulators, and safety institutes cite when they need something concrete to point at — the way aviation cites ICAO standards, the way automotive safety cites ISO 26262. Not the political authority. The engineering specification that makes the politics implementable.

The race is not to build the biggest AI. It is to build the governance architecture that makes the biggest AI trustworthy. That architecture should not belong to any one of us. It should belong to all of us.

What You Can Do Today

1. Read the standard: [compliance.sovereignsos.com/standard](/standard) 2. Download the specification (PDF): [compliance.sovereignsos.com/api/standard/whitepaper](/api/standard/whitepaper) 3. Read the patent pledge: [compliance.sovereignsos.com/standard/patent-pledge](/standard/patent-pledge) 4. Become a Founding Signatory: Sign at [compliance.sovereignsos.com/consortium](/consortium) — name, organisation, constituency, and a brief statement of endorsement. The founding cohort closes at 42. 5. Cite F-ACT in your policy, procurement, or vendor questionnaire. The standard is free to require. The more organisations that require it, the faster it becomes infrastructure.

---

"The standard that governs how autonomous agents act on behalf of humanity should itself be governed by humanity — not by any single company, government, or bloc. This is our invitation to build that together."

— Darryl S. Astin, 19 August 2026 · World Humanitarian Day