The ROL: Talking Truth to Power
Rita Felgate | June 2026
ruleoflaw.science
An em-dash publication: developed through human-AI cognitive synthesis.
A. The Crisis Is Already Live in Governance
The AI governance crisis is not waiting in the future. It is already live inside the machinery of state.
AI systems are already being used to process benefits, generate enforcement recommendations, assist border-control decisions, support policing and intelligence functions, and operate inside public-sector information systems. These are not private conveniences. When AI performs or supports a public function under state authority, it participates in the exercise of state power.
The scale of AI use is no longer speculative. Over 70% of public servants worldwide now use AI, yet only 18% say their governments are using it effectively (Public Sector AI Adoption Index 2026, Center for Data Innovation — survey of 3,335 public servants across 10 countries including South Africa). In the United States alone, 55.7% of government organisations now use AI — but only 42.9% have formal AI policies in place (2026 State of Digital Government report, Granicus). Predictive systems generate sentencing recommendations, benefits eligibility decisions, and border security assessments across multiple jurisdictions.
The governance vacuum is visible not just in what AI does in the field, but in how the terms of its deployment are determined. In 2025, Anthropic signed a $200 million contract with the US Department of Defense, stipulating that Claude could not be used for mass surveillance of US citizens or fully autonomous weapons systems. The DoD demanded unrestricted use. Anthropic refused. The Trump administration banned Anthropic from all federal contracts and designated it a national security risk. OpenAI stepped in within hours, negotiating a Pentagon deal with nominally similar restrictions. The terms of AI deployment in the most consequential governance contexts — military targeting, intelligence operations, autonomous weapons — are being set by private contract negotiation between technology companies and defence departments, with no common standard, no independent verification, and no accountability architecture that applies uniformly to any actor. That is not governance. That is transactionalism — and the outcome depends entirely on which company holds which line, and for how long.
The moratorium proposal covers frontier AI development. It says nothing about this. That silence in the face of the obvious cannot be accidental — AI is already so embedded in the machinery of state governance that any moratorium covering deployed systems would require dismantling operational state infrastructure. The UK’s Department for Work and Pensions uses AI to process Universal Credit eligibility assessments at a scale that manual staffing could no longer support. US Customs and Border Protection runs facial recognition at borders handling hundreds of thousands of daily crossings. Targeting recommendations in active conflict zones are AI-generated. Dismantling that is not a policy option any government will take, and no government has offered to.
On June 4, 2026, Anthropic — the company that makes Claude, currently among the world’s most capable AI systems — called for a global pause in frontier AI development. The moratorium proposal addresses AI development. It does not address AI already deployed in state governance. Even a successful moratorium would leave that problem entirely unresolved.
It is the Rule of Law, not a moratorium, that has the answer. It is not new. It has been sitting in governance systems for as long as governance has required the separation of powers to function.
B. AI in Governance Is an Exercise of State Power
The crucial point is simple.
When AI is used to perform, recommend, filter, prioritise, classify, target, approve, deny, monitor, or enforce within a state function, it is not merely a technical tool. It is part of the chain through which state power is exercised.
If an AI system assists in deciding eligibility for public benefits, the decision remains a governance act.
If an AI system assists border control, the resulting exclusion, delay, detention, or classification remains a governance act.
If an AI system supports policing, surveillance, military targeting, sentencing recommendations, procurement decisions, licensing, or regulatory enforcement, the state cannot avoid accountability by pointing to the instrument.
The question is not whether the actor is human or artificial. The question is whether state power is being accessed or exercised, and whether that access and exercise meet the conditions required by the Rule of Law.
Those conditions are not optional. They are what lawful coordination of separate powers requires.
Deployment of AI in state governance carries not only legal consequences, but direct rule of law consequences. Every AI-generated action taken under the umbrella of state power is an exercise of state power, and the Rule of Law has always been the accountability architecture for exactly that. Governance under the ROL is the platform through which the governance of AI will itself be resolved: the tools that make AI accountable in governance are the same tools that make governance accountable full stop.
That connection goes unrecognised in every statement made about the moratorium. The failure to recognise it is the governance crisis that sidelines the ROL as the instrument to resolve it.
C. The Two Fantasies in Heart-failure
The moratorium proposal did not emerge from nowhere. It emerged in a world seized by ROL failures that have made states vulnerable to two forms of state capture, both cast as solutions to those failures — dismantling of the state itself, and capture of state institutions — both running simultaneously, both visibly failing on their own terms, each directly producing the further instability they claim to address.
Big-gun geopolitical fantasies: the fantasy that military dominance effects peace among nations. States following big-gun geopolitical fantasies from a bygone era where might was right are discovering the resilience of states in resisting attacks on their sovereignty and independence which litter the geopolitical landscape: Russian attempts to subjugate Ukraine, USA attempts to draw the nuclear teeth of Iran, both merely resulting in a metamorphosis of the battlefields which predicts the same for other conflicts around the globe. The failure is visceral, not abstract: fields of rotting corpses, economies fracturing under the weight of conflicts that cannot be won and cannot be stopped, leaving advocates with no exit strategy and accountability for the decisions that produced those consequences nowhere to be found — because the separation of powers that would enable it has been sufficiently broken to make the decisions possible in the first place.
Metamorphosis of the state fantasy: that great wealth acquires enough technological control to direct the outcomes the fantasy describes — to dismantle state architecture, appoint sympathetic institutions, gut regulation, deploy AI in state governance functions, and emerge with a world remade according to the ideal. The AI deployed outside of the accountability architecture of the state produces results nobody fully anticipated. The institutions captured fracture into competing factions. The regulation gutted removes constraints on everyone, including those who gutted it. Real estate grabs are made on territories of other states still occupied by their traumatised populations — as if the power to designate state territories is interchangeable with the reality on the ground.
The architect of the most powerful AI in the world stands before the world and proposes a voluntary pause — the thing he could say with integrity — because the architecture that would make his position resolvable doesn’t yet exist in accessible form.
The separation of the two fantasies is for descriptive purposes only. In reality both fantasies coincide, feed off each other, with those holding the reins like Roman riders with a foot on the back of each horse. Each gave the other cover: geopolitical necessity justified dismantling oversight; technological dominance justified geopolitical adventurism. The cover story held as long as both appeared to be working. Both are now visibly failing simultaneously — and the cover story has collapsed with them.
The two fantasies are not the solutions to state failure they purport to be. They are descriptions of the form of coordination failure that emerges when ROL architecture that constrains access to and exercise of state power breaks down sufficiently for force and wealth to govern the space that ROL governance vacated. The structural consequences are severe — and both poles are now experiencing them. Poles apart in their politics. Identical in their structural position. Both asking, in the vacuum of the ROL: what now — and where to?
D. A Moratorium: An Unenforceable Gesture with No Exit Strategy
Anthropic’s moratorium proposal is the answer that emerged from that question.
The gesture names the problem with full technical understanding of what is being built and where the trajectory leads: society is not ready for AI. And proposes a self-admittedly unworkable solution: a voluntary worldwide pause in advanced AI development. The architecture that would coordinate the solution is not identified. It is a gesture without a mechanism.
Voluntary coordination cannot solve a conflict driven by competing interests. Anthropic’s moratorium proposal identifies its own fatal flaw: a voluntary slowdown might simply “allow the most reckless actors to catch up technologically,” leaving everyone worse off. This critical caveat recognises that the necessary structural conditions for voluntary cooperation do not exist. A system lacking common standards, verification, enforcement, and compliance incentives is not a coordination mechanism — it is an inevitable failure mode.
The moratorium reaches for the nearest available response rather than the structural one.
The structural contradiction is now visible in financial markets. Anthropic submitted its confidential IPO filing to the SEC on June 1, 2026 — the same week its moratorium proposal was published — at a valuation approaching a trillion dollars. OpenAI followed on June 8, valued at close to another trillion. SpaceX had filed in April. Three companies simultaneously calling for coordinated restraint in frontier AI development, simultaneously racing each other to Wall Street. Capital does not pause for voluntary agreements. It moves toward the return. The moratorium proposal and the IPO filings are not in tension — they are the same structural reality: governance architecture that cannot hold the forces it is trying to coordinate.
The structural response already exists. It lies within the ambit of the ROL.
E. The Rule of Law Supplies the Missing Standard
The Rule of Law is not one option among many. It is the accountability architecture that makes the exercise of state power lawful — and it applies whether the powerful like it or not.
In the Science-based Rule of Law framework, that architecture is expressed through a practical standard: Reason, Rationality, and Logic (RRL).
Reason requires a factual basis for the exercise of power.
Rationality requires a demonstrable connection between the power exercised and the lawful purpose for which it is used.
Logic requires consistency with the structure of the system as a whole — including separation of powers, component authority, public accountability, and the rights and duties affected by the act.
These requirements apply to legislation. They apply to executive action. They apply to judicial decisions. They apply to the Public and its conduct in relation to the state. They apply to every authorised exercise of governance power.
They also apply to AI when AI performs or supports state governance functions.
For AI deployment in state governance, the Rule of Law therefore asks:
Reason: What factual basis justifies using AI for this public function?
Rationality: Does the deployment serve the lawful purpose of the function, or does it distort, evade, or undermine it?
Logic: Is the deployment consistent with the separation of powers, the rights of the Public, the accountability of the Executive, the oversight role of the Legislature, and the adjudicative function of the Judiciary?
These are not new questions invented for AI. They are the questions governance has always had to answer. AI makes them more urgent. It does not change the standard.
F. The Seven Documentary Tests
The Science-based Rule of Law framework makes the standard operational through seven documentary tests.
For any governance instrument to be actionable, those subject to it must be able to identify it, understand it, verify that it is in force, and know what it requires. This is not a technical preference. It is the minimum condition for lawful governance.
The seven tests, applied to AI governance instruments — the statutes, regulations, licences, deployment protocols, operational standards, and published policies that states issue:
Authority: Who issued the AI deployment instrument, and under what legal power?
Clarity: Is what the instrument permits, requires, prohibits, or authorises understandable to those affected by it?
Commencement: When does the authority take effect?
Publication: Has the instrument been made publicly accessible in a permanent and retrievable form?
Public Participation: Where participation is legally required, were affected persons given a genuine opportunity to participate before the instrument was issued?
Referent: Can the legal basis, cross-references, standards, and incorporated documents be identified and verified?
Jurisdiction: Is the issuing entity acting within its authorised domain?
If the authority cannot be identified, the act fails.
If the obligation cannot be understood, the act fails.
If commencement cannot be verified, the act fails.
If the instrument cannot be accessed, the act fails.
If participation was required but absent, the act fails.
If the legal referents cannot be checked, the act fails.
If the issuing body acted outside its domain, the act fails.
That is how the Rule of Law turns anxiety about AI into a testable governance question. The question is not whether a state says it governs AI. It is whether the public record shows that it does.
In South Africa, twenty years of gazette data shows that approximately 30% of governance instruments fail at least one of these categories. That finding is not about AI governance specifically. It establishes the baseline: this is what governance failure looks like in documentary form, before AI arrived to amplify its consequences.
G. Why This Is Stronger Than a Voluntary Pause
A moratorium asks actors to slow down.
The Rule of Law asks whether they have authority to act at all, whether that authority is being exercised lawfully, and whether the limits on that authority are being enforced.
A moratorium depends on voluntary compliance.
The Rule of Law creates a common standard.
A moratorium struggles with verification.
The Rule of Law begins with documents that must already exist if public power is being lawfully exercised.
A moratorium has no obvious exit strategy.
The Rule of Law provides one: bring AI deployment under lawful authority, make that authority public, test it against objective standards, and hold non-compliant exercises of state power structurally accountable.
That is the coordination architecture the moratorium lacks.
Common standard: RRL applied to the exercise of AI deployment power. Objective. Testable. Verifiable from the documentary record — without access to proprietary systems, without trust, without inspection.
Verification: Documentary records of what states publish about AI governance are subject to the seven-category test. Any researcher, advocate, or member of the Public can apply it. The verification is open, not closed.
The Unity of States Commons: ROL-compliant states have a structural interest in other states’ AI governance compliance. Not a moral preference — a coordination interest. Because there is a common standard and objective verification, the comparative performance of states against the ROL becomes objectively verifiable. The Unity of States Commons can respond to states with increased risk profiles by reducing the benefits of participation — creating a structural incentive for compliance that no voluntary mechanism can replicate. States that meet the documentary standard participate in the commons. That is the coordination architecture the moratorium lacks and cannot supply.
H. The Public Is Already at the Gate
The Public has already recognised the problem, even if the formal vocabulary has lagged behind.
Public protest outside AI companies, defence contractors, and state institutions is not merely social anxiety about technology. It is the Public component appearing at the point where governance outputs fail to meet what governance requires.
The phrase “Rule of Law” already carries public force because it names the intuition that power must answer to a standard. The Science-based Rule of Law framework makes that intuition precise. It turns the rallying cry into a test.
The seven categories are not reserved for specialists. They are questions any person subject to governance is entitled to ask:
Who authorised this?
What does it require?
When did it take effect?
Where was it published?
Were affected persons heard?
Can its legal basis be verified?
Was the issuing body acting within its powers?
When AI is deployed in state governance, those questions are not optional. They are the Public’s entry point into accountability.
I. Talking Truth to Power
The Rule of Law does not begin by threatening powerful actors. It begins by applying a standard.
That is why it is capable of speaking to power of every kind: military, political, technological, financial, administrative, judicial, and public. It does not ask who is strong. It asks whether the exercise of power is authorised, reasoned, rational, logical, verifiable, and within jurisdiction.
Those who meet the standard gain the stability that lawful coordination provides.
Those who do not meet it face the structural consequences of non-compliance.
The Rule of Law is therefore not a moral appeal to better behaviour. It is not a plea for voluntary restraint. It is not an expression of institutional nostalgia. It is the operating architecture through which power becomes governable.
The AI moratorium debate could not name this answer because the debate was framed around development rather than governance.
But AI is already in governance.
The answer is already available.
The Rule of Law applies now.
A Note on This Post
This is the first public-facing application of the Science-based Rule of Law framework to the defining governance challenge of our time. The framework has been in development for twenty years, grounded in South African governance data and the formal scientific method. Its claim is practical: where AI participates in the exercise of state power, the Rule of Law supplies the standard by which that deployment must be authorised, verified, and held accountable.
The choice is no longer whether AI will enter governance. It already has. The question is whether governance will require AI to answer to the Rule of Law.
A note on method: this post was developed through the Em-dash methodology — human-AI cognitive synthesis between Rita Felgate and Claude (Anthropic). The analysis is critical of Anthropic’s moratorium proposal. Claude has been directed to hold a watching brief against bias toward its creator and has flagged no instances of that bias in this analysis. The structural critique stands or falls on its own terms.
Rita V. Felgate is an independent legal practitioner and governance researcher. She is the founder of ruleoflaw.science and the developer of the Science-based Rule of Law framework.
Contact: ruleoflaw.science
Leave a Reply