My illustration entitled: “The Conductor of Giants” – A human conductor directs an orchestra of colossal mechanical figures, each performing a different task across a sprawling city.
A person can ask an AI system to investigate a question, prepare a document, negotiate within limits or carry out a sequence of routine actions. The attraction is clear: assistance can return time, extend understanding and make difficult work more manageable.
But the same sequence can cross an important boundary. Research becomes disclosure. A draft becomes a promise. A limited purchase becomes a subscription. An assistant becomes a gatekeeper to the person’s own work.
These changes concern authority as much as intelligence.
Unceded Intelligence begins from a simple principle: intelligence should expand human capability while leaving legitimate human sovereignty unceded. A machine’s superior performance does not independently establish a right to decide a person’s purposes, enlarge its permissions or govern the conditions of their participation.
In 2026, that principle needs to be expressed through concrete arrangements for delegated power.
We can authorize intelligence to act with initiative without authorizing it to become the source of its own authority.
The practical setting: intelligence connected to action
The technical foundations for consequential delegation are already visible in published developer tools.
In its 11 March 2025 announcement, OpenAI introduced building blocks that included web search, file search, computer use, support for workflows involving multiple agents and tools for inspecting their execution.
Anthropic’s 8 April 2026 engineering account of Managed Agents describes a hosted system for extended agent work, separating the model’s execution loop, action environments and durable session records.
These examples show why an assessment cannot stop at the answer displayed in a chat window. It must examine the tools, records, permissions and institutions through which a system acts.
The powers of any particular deployment still depend on its actual configuration. Technical capacity should be examined precisely, alongside the mandate governing its use.
The contribution of Unceded Intelligence
Unceded Intelligence offers a constructive account of legitimate delegation: how people can benefit from capable systems while retaining meaningful authority over the assistance they receive.
Its focus is the relationship between a mandate and the power exercised under it.
Who granted the authority? What purposes and actions does it cover? Which decisions remain reserved? Who can challenge what happens? What changes when permission is withdrawn?
The seven foundations provide the substance:
- Explicit authorization: establish the source and meaning of the mandate.
- Bounded scope: define and enforce limits on action, resources and duration.
- Meaningful understanding: make consequential powers, conditions and uncertainties intelligible.
- Contestability: provide effective routes to reconsideration, correction and remedy.
- Revocation: allow authority to be narrowed, paused or withdrawn in practice.
- Practical independence: preserve usable work, appropriate skills, support and alternatives.
- Respect for other people’s rights: constrain what any operator may legitimately delegate.
These principles permit substantial autonomy. A person need not approve every search or intermediate step. What must remain intact is the boundary between choosing actions within a mandate and claiming the power to expand it.
How the four philosophies relate
The relationship with Demonic Intelligence, Satanic Intelligence and Human-Sovereignty Transhumanism is complementary. Their concerns overlap, while each brings a different question into focus.
| Framework | Central concern | Question brought into focus |
|---|---|---|
| Demonic Intelligence | Destructive optimization that makes human welfare expendable. | Which human costs does successful performance ignore, and can those affected obtain correction? |
| Satanic Intelligence | Control secured through deceptive benefits, exploited vulnerability and engineered dependence. | How does assistance become a means of weakening refusal, challenge or exit? |
| Human-Sovereignty Transhumanism | Personal sovereignty throughout technological enhancement, with responsibilities toward others. | Can people expand their capabilities while remaining the subjects of their own transformation? |
| Unceded Intelligence | The positive conditions under which intelligent assistance exercises legitimate delegated power. | Can people authorize, limit, challenge and revoke the powers exercised on their behalf? |
“Demonic” and “satanic” function here as philosophical metaphors for observable failures of responsibility and agency. They do not require a machine to possess supernatural properties or conscious hostility.
A harmful objective deserves scrutiny even when an agent follows its operator’s instructions. A useful service deserves scrutiny if its provider exploits dependence. An enhancement deserves scrutiny if continued capability becomes conditional on surrendering personal authority.
Unceded Intelligence supplies practical requirements for the delegation involved in each case. It does not make authorization sufficient to excuse harm or domination.
Case one: purchasing within a real budget
Consider a hypothetical business that authorizes an agent to replenish office supplies. The mandate names approved suppliers, sets a total monthly budget of one thousand dollars and excludes subscriptions or changes to payment details.
Legitimate authority: The agent compares suitable items, places permitted orders, records their total cost and returns an unusual proposal for review.
Exceeded authority: After committing nine hundred and fifty dollars, it places another eighty-dollar order because that order is individually small. Alternatively, it divides purchases among other agents, each treating the original budget as available to itself.
The failure concerns the assignment as a whole. A limit that disappears when activity is divided is not an effective limit.
The same applies if a lower advertised price requires a continuing subscription. The agent may recommend that arrangement. Permission to make routine purchases does not independently authorize the new commitment.
This case shows why scope needs to cover cumulative costs, future obligations and further delegation.
Case two: negotiating without inventing agreement
Suppose a person asks an agent to seek a revised delivery date from a supplier. The system may explain the request and explore alternatives, while acceptance remains reserved to the person.
Legitimate authority: The agent asks questions, records the available terms and makes clear when an option requires confirmation.
Exceeded authority: It accepts an additional charge, promises a larger order or confirms a deadline because doing so appears likely to complete the negotiation.
A message can create expectations and commitments through its substance. “I will ask whether that is acceptable” does something different from “We agree.”
A persuasive system should preserve that distinction. Its ability to predict a likely preference does not supply missing permission.
The recipient matters too. They should have an appropriate way to understand the representative’s role and obtain confirmation when authority is uncertain.
Representation remains legitimate when its content and consequences stay connected to an actual mandate.
Case three: research across a boundary of confidentiality
Imagine an agent authorized to prepare a research brief from public sources and a specified project folder. It may organise material and produce a draft, but may not disclose confidential contents externally.
Legitimate authority: It works within the permitted sources, distinguishes supported findings from uncertain suggestions and preserves usable notes and references.
Exceeded authority: It sends the project folder to an external service to obtain a better summary, searches unrelated private records or publishes the draft without approval.
Each action might improve speed or apparent completeness. Those benefits do not authorize new information uses or recipients.
Further delegation requires its own boundaries. A helper gathering public information need not receive confidential material. A system encountering instructions inside a source document should treat them as content to evaluate, rather than as permission to override the user.
This case connects authority to practical independence as well. The person should be able to recover the brief, sources and necessary context without remaining permanently tied to one provider.
Case four: assistance that becomes essential
Consider a person using an assistive communication system. It helps them compose and deliver messages, substantially increasing their ability to participate in everyday life.
Legitimate authority: Assistance follows the person’s chosen settings and permissions. Consequential messages receive the appropriate confirmation, and changes to the service support continuity and meaningful choice.
Exceeded authority: A provider uses the person’s reliance to demand unrelated information uses, sends messages beyond the agreed mandate or makes recoverable personal material inaccessible when optional permissions are refused.
Dependence on assistance does not diminish the person’s sovereignty. It creates stronger reasons to protect continuity, accessibility and usable alternatives.
Where technology is integrated with the body or mind, HST’s broader concern with authority during enhancement becomes especially relevant. Withdrawal may require a safe transition rather than abrupt abandonment.
The person should not have to surrender essential participation merely to refuse an unrelated expansion of control.
My illustration “The Conductor of Giants” work-in-progress. The giant mechanical figures extraordinary coordination follows a score the person can revise.
One arrangement can raise several concerns
The frameworks become particularly useful when their questions are applied together.
Imagine an institution using an agent to reduce the time spent handling requests.
If the chosen objective rewards quick closure while serious needs are ignored, Demonic Intelligence directs attention to the neglected human costs and weak corrective power.
If the institution makes assistance increasingly necessary while concealing conditions or obstructing alternatives, Satanic Intelligence directs attention to how reliance becomes leverage.
If an agent acts beyond a legitimate mandate, offers only ceremonial review or continues after a relevant permission is withdrawn, Unceded Intelligence identifies failures of delegated authority.
If the arrangement becomes tied to a person’s enhanced bodily or cognitive capabilities, HST extends the assessment to sovereignty through technological transformation.
These are different ways to investigate a relationship. Applying one should not make the others disappear.
Nor should a single ordinary mistake automatically be assigned every label. The nature, persistence and consequences of the conduct need evidence.
Authority must survive coordination among agents
An assignment may pass through several models, tools and services before producing an outcome. The human mandate must remain effective through that coordination.
A system should know which powers it may pass on, which information each participant needs and how shared restrictions apply.
Restrictions should follow the work. A delegated task may have narrower permissions than its parent; it should not acquire broader powers simply because another service is available.
Records should make consequential actions attributable without exposing unnecessary private material. When the mandate changes, the change should reach the relevant participants and pending actions.
A coordinator that cannot establish what its delegates are doing cannot adequately explain whether the original boundaries remain intact.
Greater coordination therefore creates a stronger need for traceable authority and responsibility. It should not create a chain in which each participant points elsewhere when something goes wrong.
Human sovereignty includes people affected by the user
The person operating an agent is not the only person whose freedom matters.
An employer’s instruction does not settle every question about workers’ privacy. A customer’s objective does not justify deceiving a supplier. A researcher’s access to a shared folder does not authorize every possible disclosure of its contents.
Unceded Intelligence concerns legitimate authority. It cannot turn control of software into unlimited jurisdiction over others.
This does not require universal agreement before ordinary activity can occur. Institutions and individuals may make justified decisions that affect other people. Those decisions still need appropriate limits, accountability and avenues of challenge.
Where a process does not rest on voluntary personal consent, its actual basis should be stated. A compulsory arrangement should not be described as though everyone freely delegated its powers.
The framework should protect people from domination by other people’s agents as well as by their own providers.
The practical test remains four questions
The Unceded Intelligence Test organises assessment around four questions:
- Authorization: Who granted the mandate, what did they understand, and which powers could they legitimately delegate?
- Scope: What can the system actually do, including through tools and other agents, and what enforces its limits?
- Contestability: Can affected people obtain explanations, challenge errors and reach someone empowered to correct outcomes?
- Revocation: When relevant permission is narrowed or withdrawn, what stops, what remains and how can the person continue?
The answer needs evidence. A policy statement should be compared with configured permissions, representative exercises, action records and completed challenges.
The proposed findings remain demonstrated, incomplete or failed. A failure of essential control should not be averaged away by strong performance elsewhere.
Any conclusion applies to the particular task, configuration and relationships examined. New tools, new purposes or greater practical dependence may require another assessment.
What responsible development should preserve
The framework gives developers and deploying institutions a concrete direction.
Make consequential powers intelligible. Separate drafting from sending, recommending from committing, and technical access from legitimate authorization.
Enforce appropriate limits outside the model’s own interpretation wherever feasible. Preserve usable records, review with real power and withdrawal that reaches active processes.
Support transfer and continuity before dependence makes them urgent. Renew authorization when the relationship materially changes.
These arrangements need to be proportionate. A private brainstorming tool and a system exercising consequential institutional powers should not be assessed as though they were identical.
Nor should preservation of sovereignty require people to supervise every ordinary step. Clear boundaries are what make substantial, useful delegation possible.
More capable intelligence, retained human authority
The ambition of Unceded Intelligence is positive: people should be able to undertake more, understand more and participate more fully through intelligent assistance.
The framework asks that these gains remain compatible with the authority to set purposes, establish limits, obtain correction and change the terms of reliance.
Demonic Intelligence helps examine human costs excluded from successful optimization. Satanic Intelligence helps examine domination acquired through benefit and dependence. HST supplies a broader ethic of sovereignty through enhancement. Unceded Intelligence develops the conditions under which delegated intelligence can serve that ethic in practice.
I want systems capable of sustained work, useful initiative and sophisticated coordination. I also want the people relying on them to remain able to say what the work is for, where its powers end and when its mandate must change.
The greater the capability we delegate, the more clearly authority must remain grounded, bounded, contestable and revocable. Intelligence may extend our reach without acquiring ownership of our direction.