MULTIPLAYER: Abdul El-Sayed and the 50/50

Published 2026-10-15 · Updated 2026-10-15 · Aidan Lawson
Tags: Inner Apocalypse · Modal Systems · Multiplayer
Mirror generated 2026-10-27T18:14:50.285Z

Abdul El-Sayed has arrived at the artificial-intelligence transition carrying an enormous bag of government.

In June, the Michigan Senate candidate proposed that the American public acquire 50% equity in the largest frontier artificial-intelligence companies. The companies would become public benefit corporations. A majority of their directors would be publicly appointed or democratically elected. Major technology companies would have to divest controlling interests. Americans would receive an annual AI dividend.

There would be independent safety testing, worker protections, restrictions on dangerous applications, and rules against allowing artificial intelligence to make certain consequential decisions.

That was AI Under Democracy.

On October 13, El-Sayed returned with AI Under Control.

Now he wants a pause on training and releasing the most powerful frontier systems, a permanent prohibition on artificial superintelligence and recursive self-improvement, an independent Cabinet-level safety agency, mandatory incident reporting, criminal penalties, enforceable shutdown powers, and an international verification arrangement with China.

The agency would inspect laboratories, block releases, and order dangerous systems stopped. Independent scientists would certify safety. A scientific board selected outside the agency would participate in its governance.

There would be emergency drills. Whistleblowers. Chip inspections. A hotline.

He has brought an entire institutional civilization to the boss fight.

And so on October 14, the Washington Post editorial board called him a Luddite.

Of course.

The editorial prefers the engineering approach recently articulated by Microsoft's Satya Nadella: keep artificial intelligence separate from the systems that authorize its actions, restrict permissions, preserve independent controls, and treat powerful models as potential insider risks.

Interesting.

• El-Sayed wants public institutions capable of stopping the machine.
• Nadella wants technical architecture capable of containing the machine.
• Modal Path Ethics would like to inspect the institutions holding the controls.

Because the really strange part of El-Sayed's proposal begins when we put his June and October plans together.

• The public would own half the company.
• The public would appoint a majority of its board.
• The public would receive some of its profits.
  • A federal regulator, acting under public law, could then order the public company's most important product shut down.

The public has become owner, regulator, customer, beneficiary, and potential claimant against the same enormous machine.

We have reached constitutional design.

Give Him the Public.

Start with what El-Sayed is actually trying to change.

Frontier artificial-intelligence companies do not operate outside the world. They depend on public infrastructure, labor, scientific institutions, energy, water, law, trained personnel, and the work of countless people who never signed their corporate charters.

Their decisions can alter the conditions under which people work, learn, communicate, obtain services, and contest institutional judgments.

Some of those consequences travel far beyond the contractual relationship between a laboratory and its customers.

A company may develop a technically excellent product while making another institution dangerously dependent on it. Its safety evaluation may be rigorous while the public has no independent way to reproduce an adverse result. An organization can be competent and still acquire powers it was never authorized to exercise.

Modal Path Ethics has spent quite some time examining this problem.

The Inner Apocalypse begins from the fact that civilization needs institutions capable of coordinating consequential work. It does not assume that all concentration is pathological. Expertise, continuity, specialized equipment, and decisive authority are often necessary.

It asks when the conditions supporting that competence begin making independent correction harder to reach.

El-Sayed's June plan directly addresses some of those conditions.

A publicly selected board could change the relation between a laboratory's management and the people affected by its decisions. Mandatory divestiture could alter concentrated platform ownership. Independent testing could create an outside capable of contradicting the developer's account.

His proposed wage insurance, expanded unemployment benefits, apprenticeships, education funding, and automation levy also recognize that technological transition does not arrive as a clean increase in national output.

People lose actual jobs.

Regions lose actual employers.

A worker cannot pay rent with a graph explaining that productivity has improved nationally.

These are real institutional concerns. Calling the entire arrangement a campaign to smash the machines does not answer them.

But public control presents a further question.

Which public?

Consider the proposed corporation.

Its directors are publicly appointed or elected. Its equity is partially owned on behalf of Americans. Its facilities consume local water and electricity. Its employees depend on their wages. Its models are used by institutions around the world.

These relations do not produce one interchangeable constituency.

• A worker may want a product delayed.
  • A patient may need the service the product currently supplies.
• A community may want the data center's cooling system replaced.
  • A national public fund may want its dividend.
• An independent laboratory may need access to the system that the corporation considers too dangerous to disclose.

All of these can be legitimate claims. They can also conflict.

The word public does not determine which claimant gets the final decision.

It does not establish that a publicly appointed director represents a particular injured worker, that American equity holders represent people outside the United States, or that a national board may decide how much water a particular watershed can lose.

A democratic appointment can constitute real authority.

The appointment still needs a function, jurisdiction, evidentiary obligations, review, and conditions under which the holder can lose.

Public ownership changes who holds valuable powers.

It does not complete the constitution of those powers.

That is our first problem.

• El-Sayed has opened the company.
  • Now the company needs a government.

Enter the Engineering.

Satya Nadella would like everyone to examine the permissions.

His October 10 essay, Models as Insider Risks in the Super Intelligence Era, makes an important distinction.

We need to separate the supply of intelligence from the authority over it.

Yes.

• A model can recommend an action
  • without possessing the credentials needed to execute it.
• An agent can identify a vulnerability
  • without receiving permission to exploit it.
• A system can propose an expenditure
  • while a separately governed process determines whether the expenditure is authorized.

The restrictions should survive the model's disagreement. They should not depend on persuading the model to obey.

Nadella proposes model diversity, observable actions, continuous verification, independent controls, independent auditing, containment, and incident disclosure. He specifically warns against allowing one model to become the sole dependency for an important result or the sole judge of its own behavior.

A system with restricted permissions, independently controlled execution, preserved evidence, and a tested shutdown mechanism can be safer than one whose developers have written an utterly magnificent constitution into the system prompt.

The Washington Post editorial prefers this approach to El-Sayed's development pause.

There is a real distinction here.

A dangerous capability may arise from the interaction of a model, tools, credentials, memory, interfaces, operators, and institutional procedures.

Preventing unauthorized action through independently enforced technical boundaries addresses that interaction directly.

A broad development restriction operates at a different level.

They may both have legitimate applications. Neither can replace the other's analysis by changing the subject.

Now follow Nadella's principle to the next institution.

Suppose the model has no authority over its tools.

Good.

Who controls the tools?

The deploying organization.

Who determines the permitted actions?

The deploying organization.

Who controls which evidence becomes available to an external complainant?

That depends on the arrangement.

Who can force the deploying organization to change its judgment when its controls are technically functioning exactly as designed?

We have left the model. We are back in the institution.

A corporate security boundary can protect the corporation's operations while failing to preserve the rights of people outside its authority. An excellent containment system can prevent unauthorized action while leaving the authorized action unjustified.

The model is denied the power to alter permissions.

The human organization retains that power.

Human control has been preserved.

Love it. Now:

What controls the human organization?

El-Sayed's regulator is one possible answer.

But the regulator presents the same constitutional problem at another level.

An agency can possess excellent scientific knowledge and still exercise authority outside its proper jurisdiction. It can classify a capability accurately while making an indefensible decision about the people affected by its restriction.

A public safety instrument does not become identical to the entire public interest because its purpose is public safety.

Likewise, an engineering control does not become a constitutional settlement because a human holds the credential.

• Nadella has correctly separated intelligence from operational authority.
• El-Sayed proposes public institutions exercising authority over the providers.
• Modal Path Ethics insists that authority itself remains answerable to other constituted powers.
  • The engineering argument is useful.
  • So is the public-law argument.
    • Neither gets to stop at the human holding the button.

The Safety Agency.

El-Sayed wants a Cabinet-level independent AI safety agency empowered to inspect laboratories, prevent releases, order shutdowns, and respond to dangerous systems.

He proposes independent testing, whistleblower protection, preserved incident logs, public breach summaries, scientific review, and emergency exercises.

Those are different mechanisms.

They deserve to remain different.

• A whistleblower may introduce evidence the laboratory would prefer to suppress.
• An independent tester may reproduce a dangerous capability.
• A regulator may possess the jurisdiction to issue an order.
• A cloud operator may possess the technical means to execute it.
• A court may later determine whether the order exceeded the agency's authority.

None of those functions follows automatically from the others.

The October plan tries to connect several of them.

Good. Now consider its strongest case.

• A frontier system has demonstrated a serious dangerous capability under independent testing.
• A narrower technical intervention cannot be shown to contain it within the available time.
  • The responsible regulator acts.
  • It orders the specified deployment suspended.
    • The provider complies.
    • Cloud operators revoke the relevant access.
    • The system stops.

An independent verifier confirms that the prohibited activity has ceased.

The stop worked.

No mysterious intelligence escaped through a maintenance tunnel.

No billionaire called in from a volcano.

No emergency commander had to negotiate with a chatbot about whether humanity deserved another afternoon.

The machinery did what its lawful operator instructed it to do.

This is a real achievement.

Now, the hospital calls.

The Hospital Would Like Its System Back, Please, Soon.

Suppose a hospital network has been using the affected provider for scheduling, forecasting demand, coordinating supplies, and maintaining complicated operational exceptions.

The hospitals did not intentionally surrender their authority.

They adopted a useful service.

Over time, staff learned its interface. Older procedures were retired. Records accumulated. Exceptional cases became embedded in operating conventions that other systems do not reproduce.

The provider's model became a significant component of the hospital network's ordinary capacity.

Then, the safety agency issued a lawful shutdown.

• The hospital's administrators agree that the dangerous capability should remain stopped.
  • They also need to keep treating patients.

Their replacement software can read much of the exported data.

It cannot perform all the work.

Some local exceptions are missing. Some staff members have never used the replacement interface. A scheduling function has no competent alternate operator. A backup exists, but its last rehearsal did not include the conditions now confronting the hospital.

None of this means the shutdown failed.

The shutdown succeeded.

The dependency was already there.

Now the public must distinguish at least three questions.

Could the dangerous capability be stopped?

Yes.

Could the lawful authority invoke the stop when it mattered?

Yes.

Can the independently protected services continue through the interruption?

Not yet established.

These are separate obligations.

The Inner Apocalypse calls the third problem functional resolvability: the maintained ability to divide, remove, replace, or survive the failure of an institution while carrying its legitimate work.

The object protected is the function that has independent grounds for continuation.

The current provider does not acquire an indefinite preservation claim because other institutions have become dependent on it.

Equally, the regulator cannot pretend the hospital's patients disappear from the field when the prohibited model stops.

A warranted shutdown can impose real local harms. Those harms may be justified by the danger prevented. Justification does not erase the remaining duties.

The response depends on the actual obligations and available alternatives.

A replacement does not need to reproduce every feature of the prohibited model. It needs to perform the particular protected work adequately within the time in which that work matters.

That takes people, equipment, records, training, permissions, funding, and practice. None of these capacities appears because a regulation says successor.

This is why the book insists upon maintained alternatives before the correction emergency. And there is a particularly uncomfortable consequence for El-Sayed's proposed public ownership.

• The hospital is depending on a company whose board is publicly controlled.
• The safety agency is exercising authority constituted for public protection.
• The public fund owns a substantial share of the company.

The public's own institutions have begun imposing obligations upon one another that none can discharge alone.

We have successfully removed the requirement that a billionaire personally authorize every important decision.

We have discovered several new people who need to answer the telephone.

The Public Wants Its Dividend.

Return to June.

El-Sayed proposes public equity and an annual AI dividend.

He also proposes an automation levy to support workers affected by technological displacement, alongside expanded benefits, training, apprenticeships, wage insurance, and small-business financing.

These are different instruments.

• A dividend distributes financial returns.
• Wage insurance protects income during employment transition.
• Training supports the acquisition of skills.
• A public service supplies some necessary function under a constituted duty.

They may reinforce one another.

They do not substitute for one another automatically.

Suppose the safety agency's shutdown reduces the public AI company's profits.

The public dividend shrinks.

This is not a reason to overturn the safety decision.

The company may have been operating a dangerous capability. Reduced profits may be an entirely acceptable consequence of preventing its use.

But if the political promise of the new system is organized around a growing flow of AI-derived revenue, safety decisions may produce conflict between the public as beneficiary and the public as protected claimant.

That conflict cannot be dissolved by identifying both sides as democratic.

The institutional roles need separate warrants.

• The public fund may own equity.
  • The safety regulator must still be capable of acting on evidence that reduces the fund's return.
• Affected workers may need support.
• The hospital's patients still need treatment.
• The data-center community may have outstanding environmental claims.
• The company may have creditors, suppliers, employees, and cleanup obligations that do not evaporate when its business model changes.

One balance sheet will not settle them all.

This is where The Inner Apocalypse moves beyond the claim that workers deserve a larger portion of the proceeds.

A person needs a functioning route to food, shelter, care, water, energy, communication, mobility, learning, and meaningful appearance before law.

A transfer of money may help secure those routes.

A payment does not establish that the necessary medicine exists, that the clinic is accessible, or that the electrical service survived the transition.

• The public can own half of a pharmaceutical company.
  • The patient still needs the medicine.
• The public can own half of an artificial-intelligence company.
  • The worker still needs somewhere to work, or a viable way to continue while the economy changes.

Retraining does not manufacture an employer at the end of the course.

An annual dividend does not become an adequate material floor because the issuing corporation is publicly owned.

El-Sayed's economic program attempts to address real displacement.

The constitutional demand is that each promised protection identify the function, actual delivery route, resources, affected people, and institution responsible when the original arrangement fails.

Otherwise the company remains the source of both the economic disruption and the promised repair.

The public may hold the shares. Its continuation remains hostage to the asset.

First, Do No Harm.

We have arrived at the October development restrictions.

El-Sayed proposes pausing the training and release of the most powerful models until enforceable safety rules and independent certification exist.

He also proposes permanently prohibiting artificial superintelligence and recursive self-improvement.

These are again distinct mechanisms.

• A conditional pause preserves a possible route to later authorization.
• A permanent prohibition closes an activity under the rules governing the ban.

Neither form is automatically illegitimate. Prohibitions can be appropriate where an activity threatens catastrophic damage, and a protective system may need to act before certainty arrives.

The question is what each rule prohibits, how its conditions are established, and who can correct the institution applying it.

Start with the language of certification.

The October platform calls for guardrails that ensure all risks to human safety are eliminated.

That is just an extraordinary condition if read literally.

All risks?

Every relevant deployment?

Every malicious use?

Every error caused by a human operator?

Every possibility of a capability escaping its stated limits?

Every harm introduced by the safety procedure itself?

A regulator cannot demonstrate the absence of every imaginable future injury.

It can specify a threat model, testing obligations, tolerated risks, prohibited consequences, evidence standards, and conditions under which authorization must be refused or withdrawn.

That would make the safety claim contestable.

• If literal elimination of all risk remains the operative requirement,
  • the conditions for ending the pause may never become satisfiable.
• If the agency instead applies a practical threshold,
  • the threshold needs an authorized source.

Otherwise the apparent absolute rule is carried by the regulator's discretion.

The regulator becomes the institution that determines when an activity is safe enough while the formal standard continues saying all risk.

That is a constitutional question hiding inside a safety adjective.

Now consider recursive self-improvement.

The June platform says AI systems used to build or improve other AI must undergo safety review, and limits should constrain agents producing agents.

The October platform calls for permanently banning recursive self-improvement, described as AI taking control of frontier AI research.

Those positions can coexist.

They require a distinction between authorized AI-assisted research and prohibited autonomous control.

Suppose researchers use a model to suggest a training optimization.

Allowed?

Suppose the model proposes a code change, a human reviews it, and an automated benchmark evaluates the result.

Allowed?

Suppose an AI research agent runs experiments inside a restricted environment while humans retain power over deployment.

Allowed?

Suppose it modifies the evaluator that determines which future experiments are authorized.

Now we have reached something constitutionally interesting.

That last case potentially changes the mechanism by which the development process judges its own work.

But the significance depends on what authority the agent possesses, what external controls remain, and whether independent correction can still reach the experiment.

A ban has to identify that relation.

Otherwise the rule risks grouping permitted assistance and prohibited control under a vocabulary that cannot reliably distinguish them.

Modal Path Ethics has already audited the superintelligence-definition problem.

We do not need to repeat that whole fight.

The remaining issue is constitutional.

A capability classification becomes a public power when it determines what other people may lawfully build, preserve, investigate, or use.

Its evidence and permitted consequences must remain contestable at the appropriate point.

• A laboratory cannot acquire immunity from safety law because its system is difficult to classify.
• A regulator cannot acquire unlimited jurisdiction because classification is difficult.

Both have work to do.

Who Watches the Watchdog?

The obvious objection arrives.

El-Sayed already proposes independent scientific governance.

His October plan calls for a board selected outside the agency, mandatory outside audits, whistleblower protection, public exemptions, incident reporting, and restrictions on officials profiting from companies they regulate.

These are serious attempts to prevent the regulator from becoming a captive of the industry.

Now inspect the arrangement.

• Who selects the scientific board?
  • Through which public authority?
• What happens when the board disputes an agency classification?
  • Can it compel independent access to the tested system?
    • Can the agency override it?
  • Does its judgment concern technical evidence, legal authorization, or both?
• If a provider contests a demand for access on legitimate security or privacy grounds, who determines the appropriate restricted route?
  • How quickly can the dispute be resolved when delay itself may cause harm?

The campaign platform does not supply a complete administrative-law design for these questions. Which is understandable.

And also exactly where implementation becomes constitutional.

The regulator needs power to act.

An emergency shutdown should not depend on obtaining unanimous permission from everyone who would suffer an inconvenience.

Some decisions must close before the threatened event arrives.

But emergency finality over one operational path does not create a general entitlement to control the evidence, future jurisdiction, appointment, review, or duration of the office itself.

An external scientific board can help preserve independence. A board with no independently usable technical capacity may be independent in appointment while dependent in judgment.

A legal right to inspect a model is insufficient if the only functioning copy, test environment, or necessary accelerator capacity remains controlled by the provider. A right to challenge an agency is similarly incomplete when no independent body can obtain the evidence or execute an authorized remedy.

The book calls for the material route through which an authority can actually lose. That route includes standing, evidence, independent judgment, operative remedy, and adequate succession where a protected function must continue.

The regulator must satisfy the test it imposes on consequential AI governance.

So must the board responsible for correcting the regulator.

And the institutions responsible for correcting that board must not all depend on the same closed record.

This does not require a regulator supervised by an infinite tower of other regulators. It requires differentiated, constituted powers whose decisions can bind without one institution controlling every path through which its own authority might later be altered.

El-Sayed's proposal opens several of those paths.

The implementing institutions would have to keep them materially available.

The Entire World.

El-Sayed also proposes an international agreement involving China.

The agreement would seek to pause and ban specified superintelligence development through verification, chip and data-center accounting, scientific cooperation, and incident communication.

Access to American chips and cloud services would be conditioned on compliance.

This takes the problem beyond the corporate board.

An international verification system needs authoritative technical findings.

• A state may dispute the finding.
• A provider may withhold information.
• A different state may have legitimate security reasons for restricting disclosure.
• An inspection can be scientifically necessary while carrying espionage or proliferation risks of its own.
• A national government may need to act against a real threat without transferring every domestic decision to an international technical body.

None of this establishes that international cooperation is futile. It establishes what the cooperation must govern.

The treaty needs specific jurisdictions, obligations, protected information, verification capabilities, dispute procedures, and consequences for noncompliance.

A technical verification institution cannot settle every competing public interest by correctly counting accelerators.

Its task is important enough to justify powerful instruments. That does not make it sovereign over every purpose for which the instruments may be used.

The same problem appears in nuclear command and control.

El-Sayed wants humans alone to decide whether nuclear weapons are used. His October plan also proposes keeping frontier AI out of early warning and decision-support systems.

Those are different functions.

• Authorization to launch is one jurisdiction.
• Detection,
  • signal interpretation,
    • spoofing analysis,
      • threat assessment,
        • and communication are others.

Suppose a bounded instrument can help detect that an apparent incoming attack is a false alarm.

Whether it should be allowed in that role depends on evidence about its reliability, failure modes, interaction with operators, and available alternatives.

The human should not lose responsibility for the decision.

But preserving human responsibility does not require pretending that every instrument near the decision exercises the same authority.

An instrument can preserve the conditions for human judgment.

It can also distort or displace that judgment.

The correct boundary has to follow the function.

A prohibition that extends across the whole warning-and-decision field therefore needs to explain which diagnostic capacities are being excluded, why, and how the protective function remains adequately carried.

Protection deserves serious institutions.

The institutions must retain the ability to recognize when their own protective apparatus has become part of the danger.

The Future Was Not at the Meeting.

One problem remains.

El-Sayed proposes permanent restrictions because some future artificial-intelligence capabilities may present dangers that present institutions cannot contain.

That is a legitimate form of concern about the future.

But the proposed restriction is itself an act taken on behalf of people who cannot yet answer.

Future researchers may discover safer architectures.

Future citizens may choose different distributions of authority.

Future institutions may obtain evidence that changes the classification of a capability.

Future artificial systems may present serious claims concerning welfare, agency, or continuity.

None of those possibilities guarantees that present restrictions are mistaken.

A future researcher does not automatically receive permission to develop a dangerous capability. A future artificial system does not receive citizenship because its producer calls it conscious. A possible moral patient does not acquire authority to expose everyone else to catastrophic risk.

The book is explicit about these distinctions.

Safety, locushood, welfare, agency, continuity, political membership, and jurisdiction are separate inquiries.

A dangerous system can be contained. A prohibited capability can remain prohibited. An institution can preserve evidence needed to investigate a serious status claim where doing so is feasible and consistent with safety.

The harder question is whether the relevant evidence and institutions remain capable of supporting later reconsideration.

The permanent ban El-Sayed proposes would initially be ordinary legislation. It would remain subject to lawful legislative revision.

The material question is what scientific capacity, records, independent research routes, and political standing would survive the prohibition.

If every serious inquiry depends on the same institution that defines the prohibited class, a future correction may exist in law while lacking the means to arrive.

If the restriction permits independent evidence preservation, lawful research within defined boundaries, challenge to an erroneous classification, and subsequent revision, the constitutional relation is different.

The future does not require advance permission to win.

It requires a real route through which new evidence and legitimate new participants can eventually change what the present has decided for them.

The Inner Apocalypse calls the underlying problem constitutional firsthood.

Founders act before their successors can participate.

They build institutions, assign powers, establish standards, and preserve some futures at the expense of others.

Their decisions are consequential because somebody has to make them.

The authority of the first participants cannot become a permanent monopoly on constitutional authorship.

A settlement that exists to protect future life must leave future life capable of participating in its correction.

The public cannot authorize itself to speak forever in the name of everyone who has not yet arrived.

That would make the original gift into an inheritance that nobody is permitted to change.

Who Can Still Lose?

Abdul El-Sayed has placed an unusually extensive institutional proposal into the AI-governance debate.

• His June plan addresses ownership, corporate governance, economic distribution, employment transition, and independent testing.
• His October plan adds restrictions on dangerous capabilities, enforceable containment, scientific oversight, incident reporting, and international coordination.

The central constitutional questions concern how those powers would be exercised together.

The Washington Post's editorial objects to heavy government intervention and turns toward Nadella's engineering argument. Nadella's proposal is valuable precisely where it distinguishes an intelligent system from the authority to act through it. And that distinction continues.

• The company may supply intelligence
  • without owning the jurisdiction of its customers.
• The public board may govern the company
  • without owning every affected person's claim.
• The safety agency may prohibit a dangerous deployment
  • without owning the conditions under which its scientific judgment can be challenged.
• The shareholder may receive a dividend
  • without acquiring authority to veto a warranted shutdown.
• The government may negotiate an international safety agreement
  • without transferring every affected institution into one technical command.
• The founder may be removed.
• The provider may be replaced.
• The regulator may be corrected.
• The scientific board may be reconstituted.
  • And the public that established the arrangement may eventually discover reasons to change it.

These are not signs that authority has failed.

They are part of what gives an authority a bounded warrant in the first place.

A society entering the age of artificial intelligence needs institutions able to make consequential decisions. Some decisions will have to be rapid. Some restrictions will have to be severe. Some dangers may justify closure before every uncertainty has been resolved.

That is why the evidence must survive. why the people carrying legitimate functions need material means of continuation, why the correcting institution must have an outside, and why no successful protection instrument can acquire permanent custody of the society whose future it was supposed to preserve.

El-Sayed's public company can have a board.

His regulator can have a stop button.

Nadella's engineers can build the controls that make it work.

Someone still has to maintain the institutions that can tell each of them when the answer was wrong.

The public may own half the machine.

It must still be able to change its mind.

Sources

• Abdul El-Sayed, First, Do No Harm: AI Under Democracy, June 29, 2026.
• Abdul El-Sayed, AI Under Control, October 13, 2026.
• Cat Zakrzewski, “Abdul El-Sayed says voters are ‘freaking out about AI.’ He wants to rein it in.”, The Washington Post, October 13, 2026.
• Satya Nadella, Models as Insider Risks in the Super Intelligence Era, October 10, 2026.
• The Washington Post Editorial Board, “The Luddite and the engineer”, October 14, 2026.
• Me, The Inner Apocalypse: A Post-Sovereign Constitution, Modal Path Ethics, 2026. Especially Chapters 4–11, 17–35, and Constitutional Articles 1–3 and 9–15.
• Me, Applied Case: The Superintelligence Ban Cannot Find the Superintelligence, September 4, 2026. The statutory-text update is live as of 10/14.

Source note: This engagement analyzes El-Sayed's published campaign proposals, not enacted law or a completed agency design. The discussion of Nadella and the Washington Post addresses their separately published arguments.