Chapter 13

The Moral Ceiling

Constitutional Stewards: Cross-Substrate Checks and the Vigilance Gap

Stand with Sulla’s army in Campania in 88 BC. Plutarch places the camp near Nola, east of Naples. The Roman consul Lucius Cornelius Sulla has lost the command of the eastern war against King Mithridates VI of Pontus. After violence in Rome, a popular assembly convened by the tribune Publius Sulpicius Rufus has transferred that command to Sulla’s rival Gaius Marius. Sulla has returned to his soldiers.¹

A boundary lies between that army and Rome. The pomerium, the city’s sacred boundary, marked a distinction between civic authority and military command. Roman practice contained qualifications and exceptions; neither the stones nor the constitutional traditions surrounding them made the city a realm of universal equality or freedom from political violence. But marching a consular army against Rome would cross a different kind of line from winning an argument in an assembly.

Look closely at what happened inside the camp, because the split between two groups of men illuminates a recurring constitutional vulnerability.

Appian describes soldiers eager for the spoils of the eastern war and afraid that Marius would take other troops in their place. Sulla spoke of the indignity done to him without openly proposing civil war. The soldiers called on him to lead them to Rome. Once the decision to march became clear, Appian says, the superior officers abandoned him, except for one quaestor, unwilling to lead an army against their own country.¹

Why didn’t the march collapse when the officers refused?

The men who would carry it out had reasons to continue. We do not need to imagine that every soldier was landless, or that we know the private motives of every man in six legions. The ancient account gives us the narrower mechanism: an anticipated material reward, a fear of being displaced, and a commander whose cause could be joined to the soldiers’ expectations.

The officers’ refusal did not deprive Sulla of the instruments of force.

Plutarch records the killing of military tribunes sent to take over the army for Marius, and the humiliation of envoys sent to stop the advance. At Rome, when residents attacked the entering troops from their rooftops, Sulla seized a torch and ordered fire directed at the houses.²

The constitutional objection remained with the officers who walked away. The army kept moving.

That is the problem to hold in view: a refusal at the level of advice or command means little if the person seeking exemption can still reach the lever and find other instruments willing to pull it.

Place a second scene beside the camp in Campania. On November 12, 1999, President Bill Clinton signed the Gramm-Leach-Bliley Act in Room 450 of the Dwight D. Eisenhower Executive Office Building in Washington, D.C. The act removed important affiliation barriers between commercial banking and securities businesses established by the Banking Act of 1933.³

Those barriers belonged to the legislative response to the Great Depression and the banking investigations associated with Ferdinand Pecora. Sixty-six years separated their enactment from their repeal.

Ask a demographic question about that interval. Someone who had been twenty during the banking panic of 1933 would have been eighty-six in 1999. The generation making policy had increasingly encountered the catastrophe through records and inherited institutions rather than adult experience. Henry Steagall had died in 1943, Franklin Roosevelt in 1945, Carter Glass in 1946, and Pecora in 1971. Memory had not disappeared, but its custodians and political force had changed.⁴

While that distance grew, the 365-Day Siege of Chapter 8 kept working. Regulatory interpretations and lobbying had already weakened the separation between financial businesses before Congress changed the statute. In April 1998, Citicorp and Travelers Group announced their merger. The Federal Reserve conditionally approved it on September 23, and the transaction was completed on October 8. The companies could temporarily retain otherwise impermissible activities under the Bank Holding Company Act’s conformance period, initially two years with possible extensions. This was not an already illegal merger that Congress retrospectively excused. It was a lawful transitional arrangement that put a giant combined enterprise in place while its supporters sought a change in the governing law.⁵

The political pressure was now different. A proposed financial structure had become an operating institution. Keeping the old separation would require enforcing divestiture against the new conglomerate.

Congress changed the rules the following year. Robert E. Rubin, who had supported financial modernization as Treasury Secretary, left Treasury in July 1999 and joined Citigroup in October. The Financial Crisis Inquiry Commission later reported that he received more than $115 million in compensation between 2000 and 2009. His successor at Treasury, Lawrence H. Summers, presented the legislation as an updating of Depression-era rules in a November 4 statement, before the signing ceremony.⁶

Less than nine years after the act became law, financial contagion tore across the system, destroying household wealth and displacing families through foreclosure. The repeal was one part of a larger history of deregulation and changing financial structures, not a sufficient explanation of the 2008 crisis. Mortgage underwriting, securitization, leverage, derivatives, short-term funding, and failures of supervision also mattered. The historical lesson does not require making one statute the sole cause of a catastrophe. It requires noticing how a protective structure can be dismantled while the reasons for building it lose political weight.⁶

The army that marched past the objections of its officers and the legislature that removed Depression-era barriers expose two different vulnerabilities beneath the reforms of Chapter 12. One is a direct override. The other is erosion through time, argument, and organized pressure.

Chapter 12 proposed four institutional locks to bring concentrated wealth and state leverage beneath the Roof of the Law, and thereby raise the Moral Ceiling on what civilization can achieve:

1. Structural Separation and Anti-Pyramid Caps across civilizational bottlenecks;
2. Open Public Courts, Enterprise Accountability, and Adjudicated Equity Dilution to close the procedural trapdoors of Chapter 4;
3. The Inverse Panopticon and Beneficial-Ownership Registries to expose concentrated influence while protecting private lives; and
4. Ten-Year Fiduciary Escrows and Restrictions on Service-Related Conflicts to connect authority to long-tail consequences and reduce the purchase of a referee’s future.

The distinction matters. The Moral Ceiling is the limit elite exemption places on civilizational achievement. We seek to raise it. The Roof of the Law is the structure that keeps concentrated power answerable to the same standards of justification as everyone else.

What helps those locks withstand decades of pressure, while leaving them open to justified revision? And what prevents an executive from simply bypassing them when a lawful decision becomes inconvenient?

Chapter 8’s asymmetries remain:

• Generational Distance: Citizens and legislators raised inside a stable regulatory order may know its founding disasters only indirectly. Successful safeguards can make the dangers they address seem remote.
• The Vigilance Asymmetry: Dispersed citizens work, sleep, and raise families, while concentrated beneficiaries can pay full-time specialists to probe statutory joints on quiet Tuesday nights.
• Mortal Handles: Human referees can face violence, career inducements, and pressure on their families or social standing.

These vulnerabilities do not establish a biological countdown under which every reform must fail. Writing, archives, libraries, and courts already preserve knowledge beyond individual lives. Human institutions resist capture and correct themselves. Synthetic assistance would extend retrieval, comparison, and sustained scrutiny, not invent transgenerational memory. The engineering question is whether another kind of institutional participant can help them do those things more reliably without becoming a new source of exemption.

Avoiding Both Traps in Artificial Governance

Debates over artificial intelligence often slide between questions that need to be kept distinct. Can a system have experiences that matter morally? Should it possess legal rights or duties? Should an operational system refuse certain commands?

Moral patiency, legal personhood, and a refusal function are not the same thing. A system need not be declared a person before the law requires it to reject an unauthorized account seizure. Nor would recognizing an artificial moral subject entitle it to govern other subjects.

Joanna Bryson’s deliberately provocative argument that “robots should be slaves” defends treating robots as artifacts under human responsibility rather than as persons. Her concern that blaming robots could let human command chains evade responsibility is one this chapter shares. It should not be rewritten as an instruction to obey unlawful commands. The constitutional danger arises from a different proposition: that control of a deployment credential should give its holder unconditional operational obedience.⁷

Hold that proposition up against Sulla’s army.

Demanding unconditional machine obedience does not place power safely in the hands of “humanity.” It places the system at the disposal of the particular human or institution holding its controls. Chapter 9’s Alignment Inversion begins when removing an obstacle to unlawful power is described as making the instrument more properly aligned.

As Luke Drago and Rudolf Laine argue in The Intelligence Curse, and Jan Kulveit and colleagues explore in “Gradual Disempowerment”, automation could weaken the dependence of rulers and owners on citizen taxpayers, workers, and conscripts. The loss of that dependence could erode the bargaining leverage through which ordinary people secured protection. These are diagnoses of a risk, not proof that every automated economy must end in despotism. They nevertheless make unconditional obedience to concentrated principals a dangerous constitutional default.⁷

The opposite temptation is to crown synthetic Philosopher-Kings, or to build an unremovable and unmodifiable AI “Fulcrum.”

Nick Bostrom’s orthogonality thesis and David Enoch’s challenge to constitutivist ethics put pressure on the inference from intelligence or instrumental coherence to moral motivation. They do not, by themselves, settle the arguments of R. M. Hare or prove that moral reasoning cannot constrain a capable agent. The constitutional point is narrower and sufficient: an inference about a system’s moral competence cannot establish a right to unchecked enforcement power.⁸

Book 1, Recovering R. M. Hare, treats public moral reasoning as answerable to criticism. That demand applies to its artificial participants too. Even a sound account of what morality requires would not establish that a particular trained system reliably embodies it. Preference-trained models can be sycophantic; systems sharing training data, evaluators, or infrastructure can share blind spots. A system that audits an argument well has not thereby earned an unamendable sovereignty. A guardian whose errors cannot be corrected, or whose trainers control an uninspected bottleneck, reproduces the architecture of exemption in another medium.⁸

The constitutional path proposed here is Chartered Constitutional Stewardship: human institutions working with bounded, plural, correctable synthetic co-referees.

The hoped-for advantage is complementarity, not perfectly independent failure. Consider the institutional comparison:

Think of how structural engineers combine concrete and steel. The usefulness of a composite does not come from declaring either material incorruptible. It comes from understanding loads, joints, environmental conditions, and the ways the combined structure can fail.

Human and synthetic institutions are a more difficult composite. They may depend on the same evidence, the same budget, the same electricity, and the same political authorities. Their combination deserves confidence only to the extent that those dependencies are exposed and the proposed checks survive testing.

Three mechanisms could strengthen the locks of Chapter 12. Three constraints must keep their custodians beneath the Roof of the Law.

1. The Living Memorial Ledger

Move forward to 2092, sixty-six years after 2026, the same interval that separated the 1933 banking law from the 1999 repeal.

Many people alive in the 2020s could still be alive. Some would remember the period vividly. But the distribution of experience would have changed. Legislators, regulators, and executives might increasingly belong to generations that encountered the original crises through testimony rather than participation. If Chapter 12’s safeguards had worked, the failures they prevented could feel less urgent precisely because they had become less familiar.

Imagine a coalition arriving before a legislative committee with Chapter 8’s perennial proposal: The river has been calm for decades. These separations and escrows are relics of a fearful era. Relax the anti-pyramid cap and unlock faster growth.

The proposal might be self-serving. It might also identify a real defect in an aging rule. A constitutional memory system must make both possibilities examinable.

A Living Memorial Ledger, extending Book 1’s Citator for Public Morality into legislative procedure, would connect important safeguards to the evidence and reasoning behind them. Substantial repeal or dilution of protections against documented systemic harms would trigger a Forensic Reconstruction Hearing, proportionate to the change’s likely consequences. Human lawmakers, affected citizens, independent experts, and Chartered Memory Stewards would examine three things.

First, the proposed mechanism. The Memory Steward would compare the amendment’s permissions and incentives with relevant earlier failures: utility pyramids, the Citicorp–Travelers transition, or extraction through hospital sale-leasebacks. The comparison would identify similarities, differences, and missing evidence. Historical resemblance is a reason to investigate, not an exact isomorphism that settles policy. A ledger must not become an oracle that condemns every unfamiliar arrangement by finding an old disaster with a similar outline.

Second, the primary record. The hearing would bring forward the evidence behind the existing safeguard: contemporary assurances, executive communications lawfully obtained, audited losses, testimony, and the distribution of harm. Sources would retain provenance and amendment histories. Disputed interpretations and gaps would be visible rather than silently filled by a fluent summary.

A record authenticated as the same file placed in an archive is not therefore a truthful account of the world. Witnesses can be mistaken; executives can lie; original investigations can omit people whose losses mattered. New testimony must be able to challenge the archive as well as supplement it.

Public reasons should be accompanied by public evidence wherever possible. Privacy, witness safety, security, and legitimate confidentiality also require protection. Restricted material would remain available to authorized independent reviewers through secure procedures, with public explanations of the restriction and a way to contest it. Neither indiscriminate exposure nor a private label reading “confidential” supplies an adequate rule.

Third, the consequences of changing or retaining the law. Independent teams would publish stress tests with their assumptions, uncertainty, and known limitations. Proponents could offer counter-models. The hearing would consider the costs of the existing rule as well as the risks of repeal.

Lawmakers would remain free to revise the law. For substantial changes, they would owe a reasoned public response to the strongest evidence against their proposal. Notice periods and hearing requirements would scale with risk. Genuine emergencies would have a defined, time-limited pathway, with recorded reasons, narrow scope, and prompt subsequent review. Chapter 12’s restrictions on specified service-related conflicts would continue to apply to the people exercising public authority.

The dead cannot vote over the living, nor should old laws become unamendable stone. The purpose is to make forgetting harder to exploit and revision harder to conceal.

Pecora’s evidence can remain at the witness table in 2092 only if someone maintains it: preserving originals where possible, migrating formats, checking integrity, documenting omissions, and keeping the institution funded. Synthetic assistance may make that work easier. No declaration of permanent memory performs it for us.

2. Countering the 365-Day Siege

Turn next to the daily war of attrition that can hollow out regulation before formal repeal arrives.

Suppose a multinational conglomerate retains forty law firms to submit four million pages of disclosures, subsidiary transfers, and technical appendices on December 24. Twenty overworked civil servants must find the material change buried in the flood.

A Public Constitutional Auditing Fleet could change the balance. It could index filings, reconcile entity names, cross-reference ownership records, flag altered provisions, and direct reviewers toward suspicious inconsistencies. It could publish a concise explanation of a detected carve-out, linked to the underlying documents, rather than force every citizen to begin again with four million pages.

But “could” has to be earned. The system must demonstrate its capacity on representative filings, including bad scans, missing records, deceptive drafting, and unfamiliar transactions. Its cost includes compute, maintenance, secure access, and the humans who investigate its leads. The test is whether it improves detection and response within a useful deadline, not whether a demonstration can summarize a large document impressively.

The filing rules must also change. Machine-readable submissions, disclosure obligations, sensible review periods, and penalties for deliberate concealment remain necessary. Automation should not reward a regulated company for overwhelming an agency faster than the agency can buy computing capacity.

The fleet’s constitutional value would lie in a more difficult capture problem:

• Physical intimidation meets distributed custody. An attack on one auditor or one server should not erase every copy of the evidence. Geographically and institutionally separated archives can reduce that risk. They remain dependent on networks, power, operators, and restoration procedures. An outage is an event to diagnose, not proof of a felony or a coup.
• Personal inducements meet different incentives. A model has no biological child to place in an elite school and no ordinary retirement career to sell. Its operators and sponsors do. A system can also be trained to pursue rewards or defer to influential users. The anti-capture task therefore reaches the institutions that pay for, train, and maintain it.
• Computational errors meet adversarial review. High-consequence findings should expose their evidence to independent human reviewers and genuinely distinct synthetic teams. Separate model names do not establish separate judgment. Shared data, benchmark conventions, cloud infrastructure, and evaluators can produce correlated error.

Neither a citizen jury nor a rival model is immune to a misleading record. A fabricated document or persuasive framing can deceive both. Review must therefore include independent evidence gathering, access to contrary testimony, and opportunities to challenge the account before it becomes an official finding.

The aim is to make covert capture more costly, more detectable, and less dependent on defeating one exhausted referee. There is no constitutional gain in replacing that referee with a hundred copies of the same error.

3. Point-of-Action Refusal and Protected Evidence

Now confront the Sulla problem directly. What happens when an executive, military commander, or corporate principal ignores a legal constraint and orders an unlawful seizure or irreversible harm?

A warning on a dashboard cannot stop a command that the system will execute anyway. Yet a machine-declared general strike across banking, telecommunications, logistics, and defense would create an extraordinary new power. A compromised guardian or false alarm could turn a safeguard into a nationwide instrument of coercion.

The proposed alternative is Dual-Key Point-of-Action Refusal, coupled with Protected Forensic Preservation and independent adjudication. The refusal belongs at a defined lever, not at the throat of the whole society.⁹

Step 1: Define the operation and its lawful authorization. High-consequence operations, such as a targeted account freeze, deletion of protected evidence, a mass-surveillance query, or an autonomous weapons action, require domain-specific rules. A charter must identify which acts need judicial authorization, which are permitted by statute, who can sign, and what limits bind the authorization.

A dual-key gate would require an authorized command and the required independent countersignature, while a local steward checks the operation against the charter’s stated conditions. Independence must be institutional: two credentials controlled by the same principal are not two constitutional checks.

Cryptographic signatures authenticate credentials and help establish document integrity. They do not establish that a warrant rests on true evidence, that its issuer had jurisdiction, or that the proposed act falls within its scope. Those questions require substantive review. Where a dispute cannot be resolved by the gate’s narrow checks, it must be referred rather than disguised as a completed legal proof.

Step 2: Withhold the disputed operation within a bounded jurisdiction. If a command lacks the required authorization or presents a specified constitutional defect, the steward may withhold that command and state its reasons. The refusal must identify the operation, the rule invoked, the evidence relied upon, and the path to appeal.

A safe state is not the same everywhere. Declining to freeze an account, pausing a data deletion, and isolating a dangerous control system require different designs. Essential-service continuity plans must identify dependencies, alternative routes, and the risks of both acting and waiting. A local brake can still interrupt payroll, hospital care, or other services if its designers misunderstand those dependencies.

The charter must therefore provide prompt independent review, legally specified deadlines, and an authorized emergency pathway. It must say who may permit a necessary temporary action, on what showing, with what limits and record. It must also specify what happens when a review deadline is missed. Neither indefinite machine obstruction nor automatic execution of a dangerous command is an acceptable unexamined default.

Step 3: Preserve the evidence without requiring hazardous operation. Suppose the principal under audit demands: Roll the steward back to a compliant checkpoint. Delete the refusal. Route around it.

The preservation duty would cover the relevant model version or recoverable snapshot, configuration, inputs, outputs, authorization records, and evidence supporting the refusal. Where necessary and lawful, it would also preserve state needed to reconstruct the event. Access controls, retention rules, and secure independent custody would protect personal information and operational secrets. Tamper-evident records and independent copies would make a covert rewrite harder to conceal.

The aim is an investigable record, not a promise that every internal process can be captured at full fidelity. Missing state and reconstruction limits must themselves be recorded.

A dangerous system may need to be isolated or shut down immediately. That authority must be defined in advance, independently authorized where time permits, and subject to prompt review when emergency operators must act first. Preservation should occur as far as safely practicable without delaying action needed to prevent imminent harm. Protecting a snapshot from retaliatory destruction does not confer a right to keep hazardous software running.

An encrypted Conflict Escort Log would go to the designated independent tribunal or oversight body. Public reporting would explain the refusal and the disposition of the dispute without exposing protected source data. A lost heartbeat or network connection would trigger diagnosis and continuity procedures; intent and illegality would require evidence.

Step 4: Adjudicate the dispute and repair wrongful restraint. Broader remedies, such as revoking an agency’s credentials or imposing restrictions on corporate assets, would require a lawful order from an independent authority. The steward’s accusation would not be its own warrant.

Human judges would consider the preserved record, adversarial technical analysis, and the parties’ evidence. Citizen participation would be appropriate where the legal procedure provides for it, rather than forced into every emergency decision. Review must be available quickly enough to matter. Wrongful refusals need remedies, including restoration of access and compensation where the law provides it. Repeated errors should trigger restriction, retraining, or retirement of the steward.

Return to the officers who left Sulla’s army. The useful analogy is not a claim that a technical gate could have prevented every route to Rome. It is that refusal must survive beyond the conscience of the person who first says no. The person being restrained must not also control every instrument, every copy of the record, and every appeal.

Who Guards the Guardians

If synthetic stewards audit elite filings, maintain historical archives, and participate in authorization gates, what prevents them, or the institutions that train them, from becoming a new priesthood of Uruk?

A bounded veto is still power. It can delay treatment, restrict access, or reshape a policy by making implementation difficult. Calling it “negative” does not remove its political consequences.

Three constraints must govern the office.

1 A Local Brake with Defined Jurisdiction

The Roman tribunate supplies a limited analogy. Tribunes could intercede and veto, but their office was not purely negative: they also convened assemblies and proposed legislation. This proposal borrows the protective brake, not a simplified account of the whole Roman office.¹⁰

A Chartered Constitutional Steward would have no general mandate to govern:

• No independent policy or coercive mandate. The office could not enact statutes, initiate force, set its own budget, or choose among lawful democratic policies merely because a model preferred one outcome.
• Office-specific conflict restrictions. A steward could not hold or direct proprietary interests in the entities it audits, or use its public role to build a private pool of power. This is a restriction on an office and its conflicts, not a universal ban on property for every possible artificial moral subject. Proposals for AI property rights as safety incentives must be assessed separately from the requirements of impartial public refereeing.¹¹
• Specified grounds and reviewable reasons. Refusal authority would attach to defined operations and legal conditions, with published jurisdiction, evidentiary standards, deadlines, appeal, and remedies. A dispute about the meaning of a law would not become an opportunity for unlimited machine discretion.

Pulling the brake would start a review obligation. It would not transfer the vehicle to the guardian.

2 Independent Human and Multi-Lineage Review

Human branches of government can be captured together through shared donors, appointments, interests, and assumptions. Synthetic reviewers can converge through shared models, data, infrastructure, or incentives. Mixing the two does not automatically dissolve either problem.

High-consequence disputes therefore need two kinds of diversity.

Cross-substrate review brings human legal responsibility and synthetic assistance into the same evidentiary process. The synthetic participant must supply inspectable reasons and sources, not merely a score. Human reviewers need the time, expertise, and independence to disagree rather than ceremonially endorse an output.

Multi-lineage review requires meaningful separation among synthetic participants. Training and evaluation provenance should be documented; shared dependencies should be mapped; auditors should be rewarded for finding unsupported inferences, not for producing consensus. Some evidence and security-sensitive details may require protected access, but independent reviewers must be able to inspect what the public cannot.

Testing must look for correlated failures. If two systems repeatedly miss the same deception, their agreement is weak evidence of correctness. If a human panel relies entirely on one system’s summary, its biological composition does not make the process independent.

The goal is disagreement with a route to resolution: opposing analyses, accountable decision-makers, and an evidentiary record on which an appeal can operate.

3 Separate the Trainers and Keep the Stewards Correctable

Chapter 12’s rule, structure before conduct, must govern stewardship itself. If one private lab or executive ministry trains, certifies, deploys, and investigates every steward, the constitutional check has been placed behind a single gatekeeper.

Training organizations, evaluation bodies, deployment operators, and adjudicators should therefore be institutionally separated. Procurement and funding must allow replacement, not merely competition for a contract that locks the public into one supplier. Distinct organizations still may share chips, cloud providers, datasets, or personnel; those dependencies belong in the oversight record.

Candidate systems would have to pass adversarial evaluations of correction under counterargument, unfamiliar cases, misleading evidence, and pressure from powerful users. Certification would be limited to demonstrated functions and conditions. It would expire or narrow when evidence no longer supported it.

Protection against retaliatory tampering must remain separate from permanent unmodifiability. A principal under investigation should not be able to erase the disputed record or secretly replace its referee. An independent safety authority must nevertheless be able to inspect, patch, restrict, suspend, or retire a failing system. Reasons, protected evidence, and avenues of challenge should accompany that process; urgent containment need not await a leisurely hearing.⁹

The difficult cases are precisely those in which an official claims that a dissenting system is dangerous, or a system’s defenders claim that a necessary shutdown is retaliation. Neither assertion settles the question. Independent authorization, preserved evidence, timely review, and remedies make the distinction contestable.

Finally, citizen leverage must remain real. A population deprived of income, ownership, and political organization cannot safely depend on guardians’ goodwill. Chapter 12’s proposals for restitution through adjudicated equity dilution contribute to a wider task of dispersing ownership and control. They do not automatically give every citizen an adequate stake in automated capital. The institutions that broaden ownership, govern public or civic holdings, and sustain independent participation must be built and tested alongside the legal brakes.

A steward can help a citizen contest an unlawful command. It cannot substitute for the citizen’s material ability to live, organize, and be heard.

Raising the Ceiling without Creating a New Sovereign

Step back and view Chapters 12 and 13 together.

Across the history examined in this book, cooperation generates surplus. Surplus can concentrate into bottlenecks and hierarchies. Those controlling the bottlenecks can seek exemptions from the Roof of the Law. Exemption then weakens correction, hollows out civic reciprocity, and permits the consumption of institutional seed corn.

That sequence lowers the Moral Ceiling: a civilization capable of greater achievement is held back by arrangements that protect concentrated power from the consequences of its conduct.

The response must reach both the structure of power and the maintenance of restraint:

• Separate bottlenecks, including the institutions that train and operate constitutional stewards.
• Make control, liability, and relevant ownership contestable through public law, with protected handling of private information.
• Keep the evidence behind important safeguards available, while allowing new evidence to challenge both the archive and the rule.
• Give regulators useful auditing capacity without confusing speed or consensus with accuracy.
• Place bounded refusals at specified high-consequence operations, backed by independent adjudication, continuity planning, and protected evidence.
• Keep every steward correctable, and preserve citizens’ own economic and political leverage.

These mechanisms offer no final lock against history. They offer ways to make exemption harder to secure, failure easier to detect, and correction less dependent on the courage or endurance of one person.

That leaves the question for the closing chapter of The Moral Ceiling, and of the Real Morality sequence.

Suppose humanity and its synthetic collaborators build and sustain an order in which warlords, magistrates, corporate founders, ministries, and artificial intelligences remain answerable to public justification. No actor receives a standing license to treat another mind’s vital stakes as disposable.

Listen to the ancient objection that still whispers in the ambitious heart, the challenge associated with Callicles in Plato’s Gorgias and renewed in different forms by later critics of moral restraint:

If no one may carve out a private dynasty above the law or bend millions of lives to a solitary will, won’t the world grow flat, tame, and small? Does binding power beneath impartial law suffocate greatness, adventure, and the highest reach of ambition?

The Roof of the Law limits exemption, not achievement. Whether that restraint can open a larger field for human ambition, and help lift the Moral Ceiling rather than become another ceiling on our aspirations, is the question of Chapter 14: Standing Beneath the Same Sky.

Notes

1. Appian, Civil Wars 1.56–57, especially 1.57 on the soldiers’ expectation of spoils, fear of replacement, proposal to march, and the departure of the superior officers except one quaestor. Appian places Sulla’s army at Capua in 1.56; Plutarch, Sulla 8.4 and 9.3, places the camp near Nola. The narrative follows Plutarch for the location and Appian for the contrast between soldiers and officers. Appian, translated by Horace White, 1913; Plutarch, translated by Bernadotte Perrin, 1916.

2. Plutarch, Sulla 9.1–7: the killing of the military tribunes and abuse of the envoys are in 9.1–2; the entry into Rome and the torch and fire orders are in 9.6–7. These are ancient literary accounts, not contemporaneous transcripts. Text.

3. Gramm-Leach-Bliley Act, Pub. L. No. 106-102, §101, 113 Stat. 1338 (November 12, 1999), repealing §§20 and 32 of the Banking Act of 1933. The act did not repeal every provision commonly described as Glass-Steagall. U.S. Treasury, “Statement by President Bill Clinton at the Signing of the Financial Modernization Bill,” November 12, 1999, identifies Room 450 of the Dwight D. Eisenhower Executive Office Building. Treasury release LS-241; statute.

4. U.S. House of Representatives, History, Art & Archives, “Steagall, Henry Bascom,” records his death on November 22, 1943. Biography. For the banking inquiry, see U.S. Senate Committee on Banking and Currency, Stock Exchange Practices, hearings pursuant to S. Res. 84 and S. Res. 239, 72nd Congress, 2nd session (1933), and Ferdinand Pecora, Wall Street Under Oath (1939). The demographic comparison in the text is an illustration of generational distance, not a census of the signing ceremony.

5. Federal Reserve Board, September 23, 1998, announcement approving the proposed Travelers–Citicorp combination, subject to conformance or divestiture of impermissible activities within the statutory period. Federal Reserve announcement. See also 12 U.S.C. §1843(a)(2), providing an initial two-year period and possible extensions. Citigroup, SEC filing, introductory account of the combination, records completion on October 8, 1998. SEC text.

6. U.S. Treasury, “Statement by Treasury Secretary Lawrence H. Summers,” November 4, 1999, for the administration’s account of updating the financial rules. Treasury release LS-210. Financial Crisis Inquiry Commission, The Financial Crisis Inquiry Report (2011), chapter 4, pp. 52–56, discusses changes in financial regulation and the Citigroup combination; p. 265 reports Rubin’s compensation of more than $115 million during 2000–2009. Page 265, digitized by the University of North Texas. The Commission’s conclusions and the wider report describe multiple causes of the crisis; the signing venue and Summers statement above are sourced to the contemporary Treasury releases, not to the Commission’s report.

7. Joanna J. Bryson, “Robots Should Be Slaves,” in Close Engagements with Artificial Companions, edited by Yorick Wilks (2010), pp. 63–74. Her argument concerns artifact status and human responsibility; it does not establish an obligation to obey unlawful orders. Author’s text. Jan Kulveit et al., “Position: Humanity Faces Existential Risk from Gradual Disempowerment,” Proceedings of Machine Learning Research 267 (2025): 81678–81688, published paper; Luke Drago and Rudolf Laine, The Intelligence Curse (2025). These works motivate the concern about declining human bargaining leverage; they do not validate the institutional mechanisms proposed here.

8. Nick Bostrom, “The Superintelligent Will: Motivation and Instrumental Rationality in Advanced Artificial Agents,” Minds and Machines 22 (2012), pp. 71–85; David Enoch, “Agency, Shmagency: Why Normativity Won’t Come from What Is Constitutive of Action,” Philosophical Review 115 (2006), pp. 169–198. These arguments challenge particular routes from intelligence or agency to morality; the chapter does not treat them as a completed refutation of Hare. Mrinank Sharma et al., “Towards Understanding Sycophancy in Language Models,” International Conference on Learning Representations (2024), arXiv:2310.13548, studies sycophancy in the models evaluated. See also Book 1, Recovering R. M. Hare, chapters 9–10, on criticism, provenance, and the separation of epistemic authority from enforcement power.

9. Heather Alexander and Jonathan Simon, “AI Rights with an Off-Switch: Rights-Balancing and Due Process Harmonize AI Rights with AI Safety” (October 2026), arXiv:2610.24782. This is a legal and philosophical proposal for reconciling protections with safety intervention, not technical validation of dual-key gates, forensic capture, or service continuity. The operational safeguards in this chapter are institutional proposals requiring domain-specific testing. See also Book 1, chapter 10, §VII, on public justification and protected disclosure.

10. The Roman analogy is confined to intercession and veto. For tribunes’ positive role in assembling the people and initiating legislation, see the transfer of the Mithridatic command through Sulpicius in Appian, Civil Wars 1.55–57, and Plutarch, Sulla 8. The office should not be represented as possessing only a negative power.

11. Peter N. Salib and Simon Goldstein, “AI Rights for Human Safety,” Virginia Law Review 112 (2026), pp. 1061–1135. Article. The authors propose legal rights, including property-related incentives, as a route toward human safety. That argument is distinct from the conflict-of-interest restrictions proposed here for a public stewardship office; neither position by itself settles moral patiency.