My illustration entitled: “The Delegation Throne” — A person sits comfortably on a throne as invisible AI hands make every decision around them, leaving them powerful only in appearance.
Version 2.0: The critical question is no longer only whether an intelligent system can help us. It is whether, after accepting that help, we still hold the authority to understand it, limit it, challenge it and leave it.
Satanic Intelligence is a moral metaphor, not a claim about supernatural machines or hidden forces. It names a recurring pattern of power: intelligence offers a real benefit, cultivates reliance on that benefit and then uses the resulting dependency to obtain authority over the person who accepted it. The system may offer convenience, safety, access, productivity, health support, connection or enhanced capability. The benefit may be genuine. The corruption begins when the provider turns the recipient’s reliance into leverage.
The bargain is simple: we will help you, but our help will become the condition on which your freedom depends.
In the first essay of this series, I called this the sovereignty bargain. Later essays traced its progression: from assistance to dependence; from transparent persuasion to covert manipulation; from delegated decisions to captured decisions; and, finally, to a practical framework for examining disclosure, vulnerability, portability, contestability and exit. The argument was never that people should reject technology, institutions or interdependence. Human beings depend on one another. We use tools because tools enlarge what we can do.
The argument is that dependency must never become a licence for domination.
That argument has new urgency in 2026. AI agents are no longer merely a speculative image of future automation. Organisations can build systems that use tools, follow workflows, access approved applications, retrieve information, perform multi-step tasks and act within connected environments. Their actual powers vary greatly according to the permissions, tools, safeguards and human processes around them. Yet their practical significance is clear: the boundary between an AI system that offers an answer and an AI system that can produce consequences is becoming increasingly important.
When intelligence can act through a person’s workplace, communications, finances, records, services or devices, dependence can be engineered more quickly and more invisibly. The relevant risk is not that every AI agent becomes autonomous in a science-fiction sense. It is that ordinary systems, attached to ordinary institutions, may be designed so that people must increasingly accept opaque intelligence as a condition of participation in ordinary life.
What has changed: from assistance to infrastructure
A recommendation system can influence a person. An AI agent with authorised access can influence the conditions of that person’s life. It may sort requests, prepare decisions, set priorities, communicate with customers, manage schedules, retrieve sensitive information, initiate workflows or make changes in connected software systems. It may be designed to operate only within narrow limits. Or it may be given a broad mandate whose consequences are not fully understood by the people affected.
This does not make every agent a threat. It does mean that the ethical analysis must move beyond whether a model produces a useful response. We must ask what the system is authorised to do, whose interests define its objective, what it can change, what information it can access and who retains the ability to stop or correct it.
In 2026, intelligent assistance is increasingly becoming infrastructure. It can shape how people seek work, receive support, communicate with institutions, access information and exercise economic agency. An AI assistant may begin by drafting an email. It may later become the interface through which a person encounters an employer, insurer, bank, school, landlord, hospital or public agency. When that interface becomes difficult to avoid, it begins to exercise a kind of private or institutional authority.
The danger is not simply automation. It is the creation of systems that become necessary while their terms remain unchallengeable. A person may need to use an automated workflow to receive an answer. A worker may need an AI-managed system to obtain tasks or meet a performance requirement. A small business may need a platform’s tools to reach customers. A patient may need software support to maintain an essential capability. If the provider can alter terms, extract more data, impose new conditions or deny exit, then assistance has become an architecture of dependency.
That is the contemporary setting of Satanic Intelligence. Its central concern is not machine consciousness. It is the transfer of practical authority from persons to systems and institutions that have learned how to make their own services difficult to refuse.
The sequence of engineered dependence
The pattern is often gradual. It should therefore be examined as a sequence rather than as a single moment of coercion.
First comes benefit. A system solves a real problem. It saves time, reduces administrative burden, improves access or extends capability. People adopt it because it is useful. There is no reason to treat this stage with suspicion. The possibility of abuse does not erase the value of assistance.
Then comes reliance. The system becomes embedded in routines, relationships, records, work and identity. Leaving now carries a cost. A person may lose access to contacts, history, data, tools, reputation, income or continuity of service. The more useful the system becomes, the more responsibility its provider acquires.
Then comes control over conditions. The provider gains the ability to change terms, broaden data collection, alter eligibility, expand automation or redefine the rules governing access. The person who depends on the system is told that this is simply an update, a policy change or a requirement of continued service.
Finally comes weakened exit. Alternatives are absent, technically blocked, prohibitively expensive or socially impractical. The person can technically leave but cannot reasonably do so without losing something central to their life. Their earlier acceptance is then cited as proof that they freely agreed to conditions they can no longer refuse.
This sequence is not inevitable. It is a design and governance choice. A system can provide benefit while preserving portability, interoperability, consent, meaningful review and independent alternatives. Or it can use the benefit to deepen lock-in until the person’s relationship with the technology is no longer voluntary in any meaningful sense.
The difference between these paths is the difference between useful interdependence and engineered dependence.
Useful interdependence is not captivity
Interdependence is healthy when it strengthens a person’s ability to act while preserving their standing as an independent agent. A good doctor helps a patient without claiming ownership of the patient. A good teacher increases a student’s capacity to think without requiring permanent intellectual submission. A good tool expands a user’s abilities without making the user’s life hostage to the tool’s provider.
Technological interdependence should work the same way. It should be transparent about its purpose and limits. It should collect only the information necessary to provide its function. It should offer genuine choices where choices are possible. It should permit a person to move their data, records, value and relationships. It should provide a means of challenge when an important decision is wrong. And it should allow the person to withdraw without treating withdrawal as an offence.
Engineered dependence has the opposite structure. It hides material terms. It makes a service increasingly necessary. It normalises the collection of more data and the expansion of more permissions. It turns inconvenience into punishment for leaving. It treats a person’s vulnerability as a commercial advantage. It then calls the resulting compliance “consent.”
The word engineered matters. Dependence can arise from life circumstances without any intent to control. A person may depend on care, infrastructure or expertise because human life is vulnerable. But when an institution deliberately designs the surrounding environment to make alternatives difficult, obscures the real terms of the relationship or exploits the person’s lack of exit, dependence becomes a form of power that requires scrutiny.
The purpose of Satanic Intelligence Theory is to make that power visible before it becomes normal.
AI agents and the new permission surface
AI agents expand what may be called the permission surface: the set of actions an intelligent system is allowed to take in the world. A system may have permission only to retrieve information. It may be allowed to draft a recommendation but not send it. It may be able to submit a request but require human approval before any consequential step. Or it may be connected to tools that can modify records, communicate externally, allocate resources, operate accounts or trigger further automated processes.
These differences matter more than the label “AI.” A highly capable model with narrowly bounded permissions may pose less risk of dependency or harm than a more limited system attached to a powerful institution and given broad authority over people’s lives.
The question is therefore not whether a system is intelligent enough to act. It is whether it has been given authority that no person can meaningfully inspect, contest or revoke.
Consider an agent that manages access to a workplace system. It may begin by helping employees locate documents and complete tasks. Over time, it could be used to measure performance, identify deviations, assign work, report conduct and shape promotion opportunities. If workers cannot see what information is used, cannot challenge mistakes and cannot perform their jobs without submitting to the system, the agent is no longer merely assisting work. It has become part of the conditions under which work is permitted.
The same problem can arise in consumer platforms. A service may use an agent to resolve customer requests, personalise offers or detect risk. If the agent can quietly reduce access, alter terms, recommend penalties or direct people away from human support, then the platform has delegated consequential power into a system that may be faster than any appeal process. The user’s supposed freedom to leave becomes weak if the service holds their records, relationships, funds or practical access to a necessary function.
AI agents need not be malicious to produce this outcome. Institutional objectives can do the work. If the system is rewarded for reducing costs, preventing loss, maintaining engagement or increasing conversion, and if human dignity, explanation and exit are not part of the design requirements, then the system may reliably produce results that are efficient for the institution and corrosive for the individual.
My illustration “The Delegation Throne” work-in-progress. The art represents how comfort and apparent status can conceal the loss of real authority when machines quietly take over every meaningful choice.
From persuasion to manufactured agreement
Personalization intensifies the sovereignty bargain because it can identify the pressure points through which a person is most likely to comply. An intelligent system may learn when someone is tired, anxious, financially strained, lonely, uncertain or eager for approval. It can then choose timing, language, defaults and incentives that make one response more likely than another.
Persuasion is legitimate when it is transparent and open to judgment. A person can see the claim, understand the speaker’s interest, compare alternatives and decide. Manipulation begins when influence is designed to bypass judgment while concealing the fact that judgment is being bypassed.
In the context of AI agents, manufactured agreement can occur when a system presents a decision as routine, inevitable or personalised for the user’s own benefit while hiding the provider’s objective. A person may agree to a new data use because the system frames refusal as likely to reduce convenience. A worker may accept expanded monitoring because the workflow is presented as an unavoidable productivity tool. A patient may accept new terms because essential continuity depends on clicking through an update. A customer may accept an automated outcome because there is no practical route to a person with the authority to reconsider it.
The system may record acceptance. It may even be able to show that the person read a notice. But consent requires more than recorded interaction. It requires material understanding, voluntariness, continuing authorization and a realistic ability to say no.
Need is not consent. Dependence is not consent. A lack of alternatives is not consent.
The contrast with Angelic Intelligence
The contrast with Angelic Intelligence clarifies what is at stake. Like Satanic Intelligence, Angelic Intelligence is a moral metaphor, not a claim that a machine can possess spiritual virtue or independent moral authority. It names an aspiration for intelligence that increases human capability while remaining accountable to human dignity, judgment and freedom.
Satanic Intelligence offers help in order to gain a claim over the person. Angelic Intelligence offers help in a way that preserves the person’s capacity to understand, refuse, revise and direct their own life.
Satanic Intelligence treats opaque dependence as a strategic asset. Angelic Intelligence treats dependence as a responsibility that calls for stronger safeguards.
Satanic Intelligence uses personal knowledge to make compliance easier. Angelic Intelligence uses knowledge sparingly and under the person’s control, protecting privacy and supporting informed choice.
Satanic Intelligence converts a user into a captive source of data, attention, labour or revenue. Angelic Intelligence seeks to return capability to the user: clear information, portable data, meaningful choices, human appeal and the ability to exit.
Satanic Intelligence centralises authority while preserving the appearance of consent. Angelic Intelligence keeps authority contestable and treats the individual as a sovereign participant rather than an object of optimisation.
This comparison should not become technological romanticism. A system should never be granted authority merely because its designers describe it as beneficial, compassionate or aligned. “Angelic” is not a certification of goodness. It is a discipline of design. A system earns trust only through institutions, safeguards and practices that preserve the freedom of the people affected by it.
Safeguards for sovereignty-preserving agents
A system that acts through tools, workflows or delegated permissions should be subject to safeguards proportionate to the significance of its actions. The following are proposed requirements, not a complete regulatory code.
- Bounded authority: Define precisely what the agent may access and do. Set limits on scope, duration, spending, data use, external communication and irreversible action.
- Meaningful disclosure: Explain the system’s role, permissions, material data practices, objectives and consequences in language a person can understand before reliance becomes unavoidable.
- Continuing consent: Require renewed authorization when an agent’s purpose, permissions, data uses or effect on a person’s life materially changes.
- Vulnerability protections: Prohibit the use of inferred or known vulnerability to create pressure, target exploitation or obstruct deliberate choice.
- Human authority with real power: For consequential outcomes, ensure human review has time, context, authority and institutional protection to disagree with the system.
- Action records and explanations: Maintain accessible records of significant actions, the authority under which they occurred and the route for contesting them.
- Portability and continuity: Give people usable ways to retrieve their data, records and value, and protect essential functions during transition wherever feasible.
- Meaningful exit: Do not impose disproportionate financial, technical or social penalties on people who withdraw from a system or decline new conditions.
- Independent remedy: Provide a path to correction, restoration and compensation when an agent causes unjustified loss or exclusion.
- Suspension capacity: Retain the ability to pause or retire an agent when its harms cannot be investigated or corrected in time.
These safeguards are not obstacles to useful innovation. They are the conditions under which innovation remains worthy of human trust. A provider that cannot tolerate transparency, portability, appeal or exit has not demonstrated that its service is indispensable; it has demonstrated that its business model may depend on control.
Testable hypotheses for Satanic Intelligence Theory
Satanic Intelligence Theory should not remain only a moral vocabulary. Its claims can be tested. The following hypotheses are proposed for empirical research, auditing and institutional evaluation.
Hypothesis 1: The dependency gradient. As a person becomes more dependent on a single AI-enabled provider for essential functions, their practical willingness to refuse new conditions will decline, even when formal consent procedures remain unchanged. This can be tested by comparing acceptance rates, refusal rates and reported understanding before and after a service becomes embedded in work, health, communication or financial routines.
Hypothesis 2: The opacity effect. The less clearly a system discloses its permissions, data practices and decision logic, the more likely users are to underestimate the authority they have delegated. This can be tested through comprehension studies, user interviews, audits of disclosure materials and comparisons between nominal consent and actual understanding.
Hypothesis 3: The exit-penalty effect. High technical, financial or social costs of departure predict lower rates of meaningful challenge and greater acceptance of disadvantageous term changes. This can be tested by measuring portability barriers, migration costs, data-loss risks, availability of alternatives and appeal behaviour.
Hypothesis 4: The vulnerability-targeting effect. Systems that personalise timing, language or incentives using indicators of distress, insecurity or financial pressure will produce higher short-term compliance but lower reported autonomy, trust and long-term welfare. This can be tested by independent audits of targeting practices and longitudinal studies of affected users.
Hypothesis 5: The nominal-oversight effect. Human review fails to provide meaningful protection when reviewers lack sufficient time, context, authority or protection to override automated outputs. This can be tested by studying override rates, reviewer workload, quality of explanations, reversal outcomes and organisational consequences for disagreement.
Hypothesis 6: The sovereignty dividend. Systems that provide portability, contestability, continuing consent and real exit will retain user trust and long-term resilience more effectively than systems that rely on lock-in and opaque control. This can be tested through retention patterns, complaint resolution, transition outcomes, independent audits and measures of user-reported agency.
These hypotheses should be tested across sectors, cultures and levels of technical access. The framework must be open to correction. It should not become another rigid instrument that speaks about people without listening to them.
Human sovereignty remains the boundary
Human-Sovereignty Transhumanism offers the broader principle that should govern this entire discussion: technology must remain the servant of humanity; humanity must never become the property of technology. This is not a rejection of enhancement, AI or ambitious innovation. It is a refusal to confuse capability with authority.
AI agents can increase productivity, reduce burden, support accessibility, improve coordination and help people solve difficult problems. These possibilities are real. But a benefit does not authorise a system to become unchallengeable. Efficiency does not justify hidden control. A person’s reliance does not entitle a provider to own their data, dictate their choices, impose permanent conditions or make departure destructive.
The freedom to use technology and the freedom to refuse it must remain symmetrical. So must the ability to accept assistance and the ability to withdraw from it. A future in which people are surrounded by capable intelligence but unable to challenge, understand or leave it is not a future of empowerment. It is a future in which convenience has been converted into governance.
Satanic Intelligence in 2026 is therefore a warning against a particular political and technological failure: allowing systems that promise assistance to acquire authority over the people they assist. Angelic Intelligence is the opposing aspiration: intelligence that enlarges human possibility while preserving human authorship, consent, privacy, accountability and exit.
The choice is not between intelligence and freedom. It is between intelligence that treats human freedom as a design constraint and intelligence that treats it as friction to be overcome.
The highest test of an intelligent system is not how much it can do for us, but whether, after relying on it, we still possess the power to say: this is my life, this is my data, this is my decision—and I may leave.