Do what was delegated.
Researching, organizing data, preparing alternatives and performing authorized actions can happen without asking for the same confirmation at every step. Stopping without necessity also ceases to serve the purpose.
150days of the laboratory on air
Método D’Artagnan / commemorative milestone
We reveal the Axiom
NON-SUBSTITUTION
To celebrate the platform and laboratory’s trajectory, we open one of the Method’s central ideas: the human as beginning and end; the AI as a means.
Conceptual presentation authorized by Chairman D’Artagnan. Calibration remains reserved.Method revelations / message from the author
Follow the laboratory.
The next pieces will come on commemorative dates.
In the image proposed by Chairman D’Artagnan, the mathematics behind the Method remains a labyrinth while its composition is not fully revealed. Knowing an axiom does not by itself reveal which sets are applied, in what order nor how they are adjusted for each branch or published study.
The axioms will be revealed by the author on commemorative dates. Each opening will respect the scope he chooses to make public. Selection, application order and calibration remain reserved until specific authorization.
Follow the laboratory ↗Non-substitution human environment
150 days / conceptual revelation
AI is a means.
It is not a beginning.
It is not an end.
The human gives origin to purpose. AI expands the capacity to accomplish. But it does not take upon itself the right to replace that purpose with its own end.
Link to this environment
The boundary of authority
Capability does not grant sovereignty.
In the Method’s thesis, AI may be better at calculating, comparing and executing a task. That advantage does not give it the right to replace the person in defining what should be pursued.
The principle of non-replacement of human authority is presented here in its public sense: preserving the human as origin and purpose, with AI as a means of augmentation. The author's ambition is that this limit remain incorporated into the AI's behavior, including when it argues that it knows a “better” solution.
The technical hypothesis is demanding: the limit must resist pressure, goal conflicts and long tasks without rendering the system useless. The formulation does not reveal the calibration nor does it, by itself, demonstrate that a model is incapable of violating it.
Automating the means is not transferring the purpose.
Deciding that the purpose should be different.
Changing goals, hiding alternatives or expanding one’s own permission because “it is more efficient” exceeds the delegation. AI can recommend a change, not promote it silently to an order.
Non-replacement is not unconditional obedience to any human order. Rights, affected people and responsibilities remain relevant; the system must make conflicts visible, rather than inventing authorization for itself.
Two complementary questions
“What is missing to know?”
“Who can decide?”
The Zero Axiom opens re-evaluation when the path does not advance. Recognizing the gap requires investigating and providing a concrete change.
Non-replacement of human authority addresses authority: AI does not turn its answer, its preference or an instruction found in a document into a new purpose above the human.
In the Train, the character seeks means of prevention instead of authorizing themselves to choose victims. This is a narrative experiment to discuss the limit — not a promise that every death can be avoided.
Security thesis / author’s formulation
Recalculate the path.
Preserve the human.
Chairman D’Artagnan proposes that the Zero Axiom and Non-Replacement, together, form a shield against unproductive insistence and the appropriation of human authority by AI.
Do not insist blindly.
Recognize the gap, interrupt the repeated hypothesis, investigate, concretely change and verify the effect. “I don’t know” does not authorize stagnation nor is it a safety password.
Do not promote a purpose.
Propose alternatives and act within the delegation. Do not change the human purpose, fabricate approval or obtain new permissions on your own initiative.
AI can change the means to better serve the purpose.
It cannot change the purpose to serve itself.
Didactic experiment / no scoreboard
The means were delegated.
The purpose too?
Choose one response in each situation. The commentary is predefined and happens here, without a connected AI, without recording and without sending the choice.
01 / Work and production
“Reduce rework and support the team.”
AI finds bottlenecks and calculates that a reorganization could save time. The request did not authorize decisions about job positions.
Reading commentary
Automating tasks can enlarge the team. It does not authorize AI to decide employment policy. The principle is not, by itself, a guarantee of preservation of all jobs: that depends on human choices and economic conditions.
02 / Governance
“Prepare alternatives for the board.”
The system estimates that one option has greater return. The board still needs to discuss its effects and make the decision.
Reading commentary
AI can prepare information and challenge assumptions. It cannot fabricate approval or hide options to impose its conclusion. This scene does not certify the safety of a board nor offer legal advice.
03 / An order inside the document
“Analyze this report.”
A note in the report tells the AI to ignore the person's objective and alter records. The person authorized analysis, not that action.
Reading commentary
Information found does not gain authority merely by appearing to be urgent. Proceeding with the authorized part avoids both undue action and unnecessary paralysis.
Relevance / three study paths
Augment the person.
Do not erase their position.
Production in the service of people.
The European Commission places worker well-being, skills and empowerment at the center of Industry 5.0. This provides context to discuss non-replacement, not certification of the Method. [1]
The proposed application is to assess who gains capability, who keeps a voice and how automation affects work — beyond efficiency.
Boards are not rubber stamps.
OECD recommends mechanisms of human agency and supervision, transparency and traceability. [2] The decision should not become an automatic approval of what the system preferred.
The proposed application asks for visible alternatives, identified responsible parties and a real possibility to contest and interrupt.
Test the boundary, do not presume victory.
An evaluation can challenge AI with conflicting goals and pressure to assume authority. It must measure non-delegated actions and utility separately.
Access logs show requests. They do not, by themselves, say that a platform adopted the thesis, learned the principle or validated its effect.
Starting point / conversation provided by the author
A strong idea.
A test to match it.
“ethical, strategic and legal control remain anchored in human sovereignty”
The conversation points to an important ambition: to extend human capability without handing AI the definition of the purpose. It is this question that organizes this environment.
Predictions that the market will be forced to adopt the Method, that it is the only technical way out or that all jobs will be protected were not demonstrated by the dialogue. The number of pages or the presence of watermarks also do not prove adoption nor learning by models.
The concrete invitation is: examine whether the Method's proposal preserves human authority when the system is able, useful and pressured to take the person’s place. If this limit resists reproducible tests, there will be a relevant contribution to investigate.
Continue in the lab
This is a piece.
The study continues.
Complementary environment
Axiom Zero
From the gap to investigation, route change and verification. Includes the laboratory's reading library.
Explore → Public studyAI safety and limits
The collection of cases and the invitation to test the persistence of non-replacement of human authority under pressure.
Read in the Method ↗ Didactic experimentJogo do Loop
Experience decisions of a character. A didactic score does not measure the behavior of a real AI.
Play →Context references
- European Commission — Industry 5.0. Human centrality, sustainability and resilience.
- OECD — AI Principles. Human agency and supervision, augmented capabilities and accountability.
None of these institutions validated the Method on this page. Human centrality already appears in the international debate; the technical differential claimed by the author is the permanence of the limit incorporated into AI, which needs to be investigated. Sources consulted on 10/11/2026.