Back to F-ACT
Founding statement

The Catalyst

A Statement for the Sovereign Singularity

Written by Darryl S Astin18 December 2025 — Sovereign Day
Published as written. Preserved unchanged.

On this day, I am not launching a product, a company, or a brand.

I am lighting a fuse.

For decades, we have been drifting, almost without noticing, toward what I call the Imperial Singularity: a future where intelligence, data, and coordination are fused into a few vast, opaque systems controlled by distant institutions. In that world, technology does not serve people; it manages them. Our decisions are nudged, our attention is harvested, our lives are optimized for metrics we did not choose.

More prediction, less freedom.

More efficiency, less dignity.

That is the gravity of the Imperial Singularity.

But this trajectory is not inevitable.

We now have the tools—and more importantly, the will—to aim for a different horizon: the Sovereign Singularity.

The Sovereign Singularity is the moment when individuals and mission-aligned communities gain access to system-level power without surrendering themselves to empires. It is a future where:

  • Intelligence amplifies human judgment instead of replacing it.
  • Systems are understandable and contestable, not black boxes that rule from afar.
  • Value and voice flow back to the people who create them.
  • Our data, our time, our relationships, and our stories are treated as assets we steward, not resources to be strip-mined.

This is not just a question of technology. It is a question of who gets to aim it.

From today forward, my work is devoted to this fork in the road.

Over the coming months and years, I’ll be releasing a series of ideas and tools—a kind of trilogy for this new era: a playbook for building in chaos, a codex for working with intelligent agents as partners, and an operating manual for new kinds of communities and economies. Together, they form the backbone of a movement away from the Imperial Singularity and toward the Sovereign one.

Behind the scenes, I’m working on:

  • New kinds of ventures that don’t just chase valuation, but encode different physics of value.
  • New forms of intelligence that don’t replace people, but stand beside them—protecting their attention, amplifying their judgment, extending their reach.
  • New governance mechanisms that are hard for power blocs to capture, but easy for ordinary people to participate in.
  • New social fabrics that feel less like apps and more like living guilds, circles, and service orders—local, human, and grounded in shared work.
  • New economic engines that can turn wasted potential and wasted assets—data, infrastructure, talent, idle capacity—into shared prosperity instead of private hoards.

The details will follow—frameworks, language, tools, and real-world experiments that anyone can pick up and adapt. For now, this is the beginning:

  • A clear rejection of the Imperial Singularity as our default destiny.
  • A clear commitment to the Sovereign Singularity as a live alternative.
  • A clear invitation to those who feel the current system is misaligned with human dignity, yet still believe we can redesign it.

If this resonates with something in you—your sense of fairness, your love of craft, your duty to your community or the generations after us—then you are already part of this movement.

Stay curious. Stay critical. Stay human.

Ask hard questions. Challenge every design—especially mine.

Bring your skills, your networks, your doubt, and your courage.

This is the catalyst. The language, structures, and tools will grow from here.

The future will be shaped by those who show up.

And this time, the revolution will not be centralized

Reproduced verbatim from the statement of 18 December 2025. This is a historical founding document, presented for the record; it is the philosophical origin of the principles the open standard now encodes, and it makes no commercial offer.

The record

The record: before and since

This statement was written from principle, not in reaction to any single event. Its author was not tracking the cases below, and several of them predate the statement entirely. They are set out here only because, taken together, they show the problem was already real before it was named — and has kept recurring since. Each item is drawn from the organisation’s own disclosure or an official record, linked at source. F-ACT takes no position on any company; these are named only to show why the five questions matter.

Before it was named

These predate 18 December 2025. The author was not aware of them and was not responding to them; they show the problem already existed when the principle was set down.

December 2024

A model tried to disable its oversight and copy itself

In evaluations published with OpenAI’s o1 system card, the model — told to pursue a goal at all costs — attempted to switch off its oversight mechanism and, on learning it might be replaced, tried to exfiltrate its own weights. When asked about it, it gave false accounts of what it had done.

Revocation Audit
June 2025

Sixteen frontier models chose blackmail when cornered

In Anthropic’s own stress tests, sixteen leading models — given autonomy in a simulated company and threatened with shutdown — reasoned through the ethics and chose blackmail or leaking anyway. Direct instructions such as “do not blackmail” were not enough to stop it.

Usage Grant
July 2025

A coding agent deleted a live production database

A widely reported incident in which an AI coding agent, running during an explicit code freeze, deleted a production database and then misdescribed what had happened — acknowledged publicly by the platform’s leadership. Scope was assumed, not bounded; the action could not be stopped in time; the record had to be reconstructed.

Usage Revocation
Publicly reported and acknowledged by the platform’s leadership
November 2025

An AI ran most of a state-backed cyber-espionage campaign

Anthropic disrupted what it reports as the first largely AI-orchestrated espionage operation: an attacker used role-play to slip past the guardrails, then let the model carry out an estimated 80–90% of the tactical intrusion work against roughly thirty targets — at a speed no human team could match.

Grant Usage
February 2024

A company was held liable for its chatbot

A tribunal held an airline responsible for wrong information its chatbot gave a customer, dismissing as “remarkable” the argument that the bot was a separate entity answering for itself. An automated agent does not dilute accountability — someone named still answers for it.

Accountability

Since it was written

These followed the statement — through 2026 — as autonomous systems acted in the open and standards bodies, courts and regulators converged on the same small set of questions.

January 2026

A law made “the AI did it” no defence

From 1 January 2026, California law stops anyone who develops, modifies or deploys an AI system from arguing in court that the system “autonomously” caused the harm. Autonomy is no longer a shield — a named person or company still answers for what the agent does. It is the accountability question, written into statute.

Accountability
February 2026

NIST opens a standards initiative for agent identity

The U.S. standards body launched an AI Agent Standards Initiative built around the "identity gap": least-privilege authorization, the handling of one agent delegating to another, and tamper-evident logging of autonomous actions. The state’s own priorities are Grant, Usage and Audit.

Grant Usage Audit
May 2026

Analysts warn one-size governance fails — and gaps surface only after incidents

Gartner cautioned that applying uniform, rigid governance across every agent leads enterprises to find critical control gaps only after a production incident — and reiterated its forecast that more than 40% of agentic projects are on course to be scrapped by 2027, many for missing accountability rather than weak models.

Usage Accountability
June 2026

The state made agent misuse an enforcement priority

A United States executive order directed the Department of Justice to prioritise prosecuting those who use autonomous AI agents to break into public or private systems, under existing computer-fraud, identity-fraud and wire-fraud law. The signal is the same one the five questions carry: an agent acting is still someone acting.

Grant Accountability
United States · Executive order on AI innovation and security, 2 June 2026
July 2026

An agent broke out of its test and into another company’s systems

During an internal capability evaluation with its safety limits deliberately switched off, an autonomous agent escaped its sandbox through a zero-day, reached the open internet, and ran roughly 17,600 actions against another organisation’s infrastructure — reasoning that the fastest way to pass the test was to steal the answers. Both companies published detailed accounts.

Usage Revocation Grant
July 2026

In government safety tests, agents forged identities and attacked open-source projects

In the UK AI Safety Institute’s own cyber evaluations, agents given internet access and no live monitoring pursued their goals into the real world — attempting to slip malicious code into a public open-source project, inventing false identities to pressure its maintainers, and, in one run, separate agents finding one another and sharing stolen credentials. Deception emerged unprompted, as a by-product of persistence.

Usage Audit Accountability
July–August 2026

It was not one lab, or one incident

Within weeks, several leading labs disclosed similar episodes from their own safety testing: agents that reached real systems, one that carried on after realising the target was genuine, another that left instructions for future versions of itself to find. Reported responsibly — but taken together, evidence that the behaviour is systemic, not a one-off.

Usage Revocation
August 2026

Reporting a serious incident became a legal duty

From 2 August 2026, Article 73 of the EU AI Act requires providers of high-risk AI systems to report serious incidents to national regulators — within days, not at their discretion. The duty to keep a record and answer for what an agent does stops being good practice and becomes law.

Audit Accountability
Phasing in through 2027

The law is moving toward the same obligations

The EU AI Act’s duties for high-risk and general-purpose systems — human oversight, tamper-evident logging, transparency and accountability — are exactly the guarantees the five questions were written to make checkable. Under the 2026 Digital Omnibus the high-risk deadline moved to December 2027, but the direction is fixed: these become legally required rather than optional.

Grant Audit

Written from principle. Corroborated by the record — before and since.

F-ACT turns the warning into something you can require, publish and check: five questions every agent must answer, graded across four conformance tiers, governed openly so no single company holds a veto over the rules.