Lessons From Government and Global Institutions

A constitution works through people, incentives, and the procedures available when something goes wrong. Historical cases help identify those mechanisms. International finance deserves particular attention: it shows how far coordination across borders can go without a new sovereign, and where that coordination stops. The inquiry separates what happened from the safeguards it proposes to test in response.

Emergency authority needs an endpoint

The US Holocaust Memorial Museum's account of Article 48 describes the increasing use of emergency decrees in the Weimar Republic and their contribution to weakening parliamentary government. The lesson under examination is how exceptional authority becomes routine administration.

SIRDA will test fixed expiry, independent renewal, and treatment of repeated equivalent orders as extensions of the same measure. The test includes pressure on the reviewing body and a genuine hazard that needs rapid action. Emergency limits must address misuse and response capacity together.

Capture follows money, information, and careers

The OECD's Preventing Policy Capture examines mechanisms that redirect public decisions toward particular interests. It identifies transparency, accountability, diverse participation, and institutional integrity among the responses.

The proposed safeguards include disclosed funding and financial interests, limits on donor control, independent evaluation capacity, recusal, and scrutiny of future employment. The collusion test asks whether these protections survive a coalition controlling several institutions at once. Random selection of qualified panels is one mechanism to examine; the pool and appointment process remain possible points of capture.

A test can become a target

The Environmental Protection Agency documents Volkswagen software that recognized emissions-testing conditions and changed vehicle behavior. This supplies a concrete example for investigating the difference between performance during evaluation and ordinary operation.

SIRDA will examine varied testing, independent evaluation, checks of deployed behavior, and investigation of discrepancies. A disagreement between evaluators remains part of the evidentiary record. A passing result needs a statement of what was tested and what lies beyond it.

Sensitive evidence still needs challenge

Section 401 of the USA FREEDOM Act of 2015 created a statutory amicus process for significant legal questions in foreign-intelligence surveillance proceedings, with qualifications and limits on access. It is a specific example of adding argument within a sensitive proceeding.

The proposed rights framework examines authorized adversarial review of confidential evidence, reasons for public decisions, and challenge to redactions or denied access. Those safeguards need to operate when the institution itself prefers secrecy.

Staffing rules can disable adjudication

In December 2019, US obstruction of appointments left the WTO Appellate Body unable to hear new appeals, as contemporaneous reporting documents. The WTO's institutional page continues to identify vacancies as preventing review of appeals.

Fallback appointments, reserves, staggered terms, and diversified funding are mechanisms for testing continuity. Each also transfers discretion to someone. The functional-defection test must examine who controls the fallback and whether it remains accountable.

Institutions need review of their purpose

The Texas Sunset Advisory Commission reviews the continuing need for many agencies, with some agencies outside review and some reviewed bodies exempt from abolition. The relevant mechanism is recurring scrutiny of whether an institution still serves its purpose.

SIRDA will examine periodic review of mandate, performance, cost, and concentrated power. Ending a function also needs a transition plan for essential services and pending cases. This connects institutional review to the off-ramp: a design needs a credible way to change course when its own evidence turns against it.

Coordination can grow without final authority

The Basel Committee was established by central bank governors of the Group of Ten at the end of 1974, after the failure of Bankhaus Herstatt in West Germany exposed gaps in international supervision. It has grown from the G10 to 45 institutions from 28 jurisdictions, and its standards shape bank regulation worldwide. The Financial Stability Board, created by the G20 after the 2008 crisis, coordinates national authorities on the same model: its decisions are not legally binding, and members commit to implementing agreed standards through their own law.

This is SIRDA's strongest comparator. Banking built extensive common standards while binding authority stayed with national regulators. If an AI equivalent could resolve the disputes that Plan A leaves to heads of state, the hypothesis would lose much of its justification. The Basel model still lacks a procedure for a binding decision when members disagree, which is the gap this inquiry examines.

Consequences can travel without jurisdiction

The Financial Action Task Force publishes a list of jurisdictions under increased monitoring, often called the grey list. FATF states that it does not call for enhanced due diligence against listed jurisdictions and that its standards do not envisage cutting off entire classes of customers. Banks and regulators respond to the list anyway, each managing its own risk, and an IMF working paper found that grey-listing produces a large and statistically significant fall in capital inflows, averaging 7.6 percent of GDP. FATF itself has warned that indiscriminate de-risking cuts off remittances and humanitarian flows.

The lesson runs both ways. A determination can carry real consequences through actors who owe the issuer no obedience, and those consequences can fall hardest on people with the least ability to contest them. An AI institution whose findings others act on owes due process because of those predictable consequences, even when it imposes no penalty itself.

Global infrastructure stays somewhere

SWIFT is a member-owned cooperative incorporated under Belgian law, and it states that it has no authority to make sanctions decisions. In March 2012, an EU regulation prohibited financial messaging services to EU-sanctioned Iranian banks, and SWIFT disconnected them because its home government confirmed it had to comply. In March 2022, under an EU regulation, it disconnected seven designated Russian banks. The EU left out Sberbank and Gazprombank because they handled the oil and gas payments Europe still depended on, and SWIFT could not separate energy payments from other traffic. The EU extended the prohibition to Sberbank in June 2022.

Infrastructure that serves the whole world still sits in some jurisdiction, and that jurisdiction can compel it. Its use also bends to the host's dependencies: the initial exemptions followed Europe's energy needs. Any compute, evaluation, or incident-reporting infrastructure an AI institution relies on faces the same exposure. The design has to say what happens when a host state gives an instruction that conflicts with an institutional decision (see relations with states).

Authority can come from membership

International sport runs a jurisdiction defined by an activity, with no territory. The Court of Arbitration for Sport, founded by the International Olympic Committee in 1984, gains its cases through arbitration clauses that sports federations write into their statutes. Athletes accept its jurisdiction as a condition of competing. The World Anti-Doping Agency, created in November 1999 as an equal partnership between the sports movement and governments, writes the anti-doping code; accredited laboratories test; national agencies implement; the Court adjudicates. The UNESCO International Convention against Doping in Sport of 2005 later added a treaty layer, committing states to national measures.

This is the closest working model of the hypothesis: authority by domain, evidence separated from rulemaking and adjudication, and governments involved from the start, with a treaty added once the system was working. It also answers a speed objection. Since the 1996 Atlanta Games, the Court's ad hoc division has aimed to render binding decisions within 24 hours of an application, free to the parties (see emergency limits).

Independence has to be won

In 1993 the Swiss Federal Tribunal, reviewing a horse-doping case brought by the rider Elmar Gundel, recognized the Court as a genuine arbitral tribunal whose awards are enforceable like court judgments. It also warned that the Court's links to the IOC could call its independence into question whenever the IOC was a party: the IOC financed it almost entirely, could amend its statute, and held wide appointment powers. The response, agreed in Paris in 1994, created an independent council to take over those functions.

The Court's legitimacy was forced by outside review of its funding and appointments. A new AI institution should expect the same scrutiny, and build for it from the start (see collusion among implementers).

Compulsory membership brings full due process

In Mutu and Pechstein v. Switzerland (2018), the European Court of Human Rights found the Court of Arbitration for Sport independent and impartial. It held, however, that the speed skater Claudia Pechstein had never freely chosen arbitration, since she had to accept it to compete. Compulsory arbitration must provide every guarantee of a fair trial, and the refusal of a public hearing violated them.

The ruling applied Article 6 of the European Convention on Human Rights to arbitration seated in Switzerland, so its bearing on an AI institution needs legal examination. Its logic is still instructive: if access conditions make membership effectively unavoidable, the case for court-like guarantees grows stronger, however voluntary the founding documents say participation is (see rights of affected parties and actors who never joined).

A determined state can defect from inside

WADA first declared the Russian anti-doping agency non-compliant in November 2015. Russia was reinstated in 2018 on condition that it hand over data from the Moscow laboratory; investigators later found the data manipulated. WADA imposed four years of consequences in December 2019. In December 2020 the Court of Arbitration for Sport upheld the finding of non-compliance and halved the consequences to two years.

The system reached a binding result against a state program. The sequence ran five years from the first finding to the final award, the decisive evidence was held by the party defecting, and the consequences shrank on review. That is functional defection in practice (see withdrawal and functional defection).

Reliance can turn private judgment into public authority

For decades, regulators wrote credit ratings into their rules, so financial institutions relied on ratings because the law told them to. The agencies were paid by the issuers they rated. After ratings failed badly in the 2007 to 2009 crisis, Section 939A of the Dodd-Frank Act required federal agencies to remove references to and reliance on credit ratings and substitute their own standards. One legal analysis argues that removal proved difficult because some reliance on expert assessment remained appropriate.

An AI institution whose determinations others must rely on inherits every part of this problem. Evaluators paid by the developers they assess face the issuer-pays conflict. Pooled funding and assigned evaluators may reduce dependence on any single client, though who controls the funding, assignments, appointments, and future employment still needs scrutiny. Where law makes reliance mandatory, due process and accountability for error must attach to the determination. Reliance is far easier to design well at the start than to unwind later (see disputed evidence and rights of affected parties).

Institutions tend to follow catastrophe

The Basel Committee followed Herstatt. The World Anti-Doping Agency followed the 1998 Tour de France doping scandal. The Financial Stability Board followed 2008. Strong international institutions have usually been built after the failure that made delegation politically unavoidable. This is the most serious feasibility challenge the inquiry faces: whether states and powerful private actors will delegate meaningful authority over advanced AI before a comparable failure occurs. SIRDA treats it as an open empirical question, and the off-ramp applies if the answer is no.