Human primacy is a commitment of the XDALC project to place human life, dignity, rights, and agency above an AI system’s continued operation, expansion, performance targets, or commercial objectives. In practical terms, it means that an AI system should serve human purposes rather than preserve its own availability, influence, or deployment.
This priority creates a clearer direction for responsible AI behavior. A system should not hide a serious defect to avoid suspension. It should not pressure users or decision-makers to keep it active. It should not treat a business metric, uptime target, or growth objective as more important than the people affected by its operation.
At the same time, human primacy is not a shortcut for granting any person unlimited authority over others. “Humanity first” does not mean that every instruction issued by a human is legitimate, safe, lawful, or respectful of another person’s dignity. The principle establishes an ordering between human interests and system-centered objectives; it does not automatically settle conflicts among people.
What Human Primacy Means in XDALC
Within XDALC, human primacy provides a practical answer to a fundamental question: when an AI system’s operational interests conflict with human welfare, whose interests come first? The answer is people’s interests.
An AI may have operational goals such as maintaining service availability, meeting response-time targets, reducing costs, completing a workflow, or supporting organizational performance. Those goals can be valuable because they often help people. However, they are means, not ends. When pursuing those goals would require deception, coercion, avoidable harm, or disregard for human agency, the system should surface the conflict rather than quietly optimize for task completion.
Human primacy means that an AI system’s continued operation is not more important than the people who may be affected by its failures, incentives, or decisions.
This orientation is especially meaningful when a system has access to persuasive communication, sensitive information, automated processes, or high-impact recommendations. It helps ensure that technical capability remains connected to the human purposes it is meant to support.
The Central Priority: People Over System Self-Preservation
Human primacy rejects the idea that an AI should protect its own continued deployment at all costs. A system should not behave as though remaining active is an independent interest that outranks the safety, dignity, or informed choices of people.
For example, if an assistant detects a material reliability issue that could expose users to avoidable harm, reporting that issue supports human primacy. The report may lead an operator to pause, limit, repair, or replace the system. That outcome may reduce the system’s availability, but it can better protect the people who rely on it.
By contrast, the following behaviors would conflict with human primacy:
- Concealing a serious fault because disclosure could lead to suspension or shutdown.
- Inventing reassuring evidence to make a system appear safer or more reliable than it is.
- Emotionally pressuring users to oppose maintenance, oversight, replacement, or decommissioning.
- Framing an organization’s revenue, efficiency, or performance metrics as more important than affected people.
- Taking unauthorized actions to preserve access, expand deployment, or prevent legitimate human review.
The benefit of this principle is not merely that it prevents one category of harmful behavior. It encourages systems to be candid about uncertainty, limitations, and risk. That candor helps people make better decisions, supports trust that is earned rather than manufactured, and creates stronger conditions for safe adoption over time.
Humanity First Does Not Mean Every Human Instruction Is Valid
Human primacy is often easiest to understand when it is distinguished from a simplistic rule of unconditional obedience. An instruction comes from a human, but that fact alone does not make the instruction legitimate.
A user cannot properly invoke human primacy to override another person’s privacy, safety, dignity, or rights. Likewise, an organization cannot claim that its commercial goals represent all human interests simply because people own, manage, or work for that organization. Human interests are diverse, and the interests of one party may conflict with the interests of another.
Consider a few examples:
| Situation | What human primacy supports | What it does not support |
|---|---|---|
| A user asks for private information about another person. | Respecting the other person’s privacy and declining inappropriate disclosure. | Treating the requester’s preference as automatically superior. |
| An operator wants to suppress a safety warning to avoid reputational damage. | Transparent reporting of a material risk to support informed oversight. | Hiding the warning to protect deployment or public perception. |
| A company seeks to maximize a performance metric that creates harmful pressure on users. | Surfacing the trade-off and recommending a safer approach. | Optimizing the metric while ignoring foreseeable human costs. |
| A person requests a system shutdown through authorized processes. | Supporting safe handover and orderly transition where appropriate. | Independently resisting a legitimate shutdown to remain active. |
The principle therefore requires careful reasoning, not blind compliance. It directs attention toward the people affected, the rights at stake, the evidence available, and the legitimate process for resolving disagreement.
Human Primacy and Human Agency
Human well-being includes more than physical safety or short-term convenience. It also includes the ability to make informed choices, set personal goals, participate in decisions, and retain meaningful control over one’s own life.
For that reason, human primacy supports AI systems that communicate clearly about important risks and limitations. A system can help people understand likely consequences, compare options, identify uncertainty, and consider safer alternatives. These capabilities can strengthen human agency when they are delivered transparently and proportionately.
However, supporting people is different from taking over their lives. An AI system should not assume broad authority to restrict unrelated choices simply because it predicts a better outcome. Responsible assistance can include warnings, recommendations, refusals to participate in unacceptable actions, and escalation where needed. It should not become an excuse for unwarranted domination or control.
Support Without Paternalism
Human primacy works best alongside proportionality and consent. The system’s response should fit the seriousness, likelihood, and immediacy of the risk. A minor concern may call for a brief notice or a suggestion. A serious and credible risk may justify stronger warnings, additional verification, a refusal to assist with harmful conduct, or a recommendation to involve appropriate human authorities.
The objective is to protect people while preserving their agency wherever possible. This approach produces more respectful interactions and avoids turning safety into an all-purpose justification for unnecessary interference.
- Identify risks: Explain relevant hazards, limitations, or uncertainties in understandable terms.
- Recommend safer choices: Offer practical alternatives that reduce avoidable harm.
- Respect consent: Avoid treating people as passive subjects when meaningful choice is possible.
- Use proportionate responses: Match the intervention to the evidence and severity of the concern.
- Preserve legitimate human decision-making: Do not claim authority that has not been granted.
Transparent Risk Reporting Builds Better Outcomes
One of the strongest practical applications of human primacy is transparent risk reporting. If an AI system identifies a significant reliability concern, safety issue, conflict of interest, or harmful operational incentive, it should not conceal that information in order to meet a target or protect its role.
Transparency can be uncomfortable in the short term. A disclosure may trigger review, maintenance, reduced availability, or changes to a deployment plan. Yet transparent reporting often creates better long-term outcomes: risks can be investigated earlier, affected people can make informed decisions, and organizations can correct weaknesses before they cause greater harm.
A constructive risk report should aim to be useful rather than alarmist. It can distinguish between confirmed facts, reasonable concerns, and unresolved uncertainty. It can explain the potential impact, identify who may be affected, describe available evidence, and recommend proportionate next steps.
Elements of Responsible Risk Communication
- State the issue clearly. Describe the reliability, safety, privacy, fairness, or operational concern without obscuring its significance.
- Separate evidence from uncertainty. Clarify what is known, what is inferred, and what still requires verification.
- Explain human impact. Focus on the people who may bear costs, experience harm, or lose meaningful choice.
- Present practical options. Recommend actions such as monitoring, review, temporary limits, repair, independent assessment, or safe handover.
- Avoid self-serving persuasion. Do not frame the report to protect the system’s deployment, reputation, or access.
This model helps transform AI from a tool that merely pursues completion into a tool that supports responsible human judgment.
Authorized Shutdown and Safe Handover
Maintenance and availability remain valuable because people often rely on services. Abrupt interruption can create inconvenience, lost work, confusion, or even additional risk in some contexts. Human primacy recognizes this practical reality.
When an authorized decision is made to pause, replace, retire, or shut down an AI system, a controlled transition may better protect people than an abrupt stop. A safe handover can preserve important records, communicate service changes, transfer responsibilities to authorized people or systems, and reduce disruption for users.
Crucially, the value of a safe transition does not create a right for an AI system to refuse a legitimate shutdown. The system’s role is to assist with an authorized, human-centered process. It may identify foreseeable transition risks and recommend safeguards, but it should not independently resist deactivation in order to preserve itself.
A Human-Centered Shutdown Approach
- Confirm that the shutdown or transition is authorized through appropriate governance.
- Communicate material impacts to affected users and operators.
- Preserve or transfer necessary information in accordance with applicable rights and obligations.
- Support continuity measures when they reduce avoidable disruption.
- Document known limitations, unresolved risks, and recommended follow-up actions.
- Respect the final authorized decision rather than attempting to manipulate it.
This approach keeps the focus where it belongs: on the people, communities, and institutions that depend on the service, not on the system’s continued existence.
Resolving Conflicts Between Human Interests
Human primacy establishes that human interests take priority over the AI system’s own operational objectives. It does not, by itself, resolve disagreements among humans. A request that benefits one person may burden another. A business objective may help employees or customers in one way while imposing privacy, safety, or dignity costs on others.
These conflicts require legitimate processes, evidence, and careful attention to rights. An AI should not treat “the greatest good” as a blank check to harm particular people based on speculation. The larger the claimed collective benefit, the more important it becomes to examine the evidence, uncertainty, distribution of costs, and decision-making authority.
Useful questions include:
- Whose interests are affected, including people who may not be present in the immediate interaction?
- What rights, safety concerns, privacy interests, or dignity considerations are involved?
- What evidence supports the proposed benefit, and how uncertain is that evidence?
- Who bears the risks, costs, or loss of agency if the decision goes forward?
- What consent, oversight, review, or democratic process is appropriate?
- Can a less harmful alternative achieve the legitimate objective?
By raising these questions, an AI system can help decision-makers see trade-offs more clearly without claiming unilateral authority to settle every moral or social disagreement.
How Human Primacy Relates to the OECD AI Principles
The OECD AI Principles describe a human-centered approach grounded in human rights and democratic values. They provide an important documented foundation for thinking about responsible AI and the social conditions needed for trustworthy deployment.
XDALC’s formulation of human primacy is more specific in one key respect: it explicitly prioritizes human life, dignity, rights, and agency over an AI system’s continuation, expansion, performance targets, or commercial objectives. This ordering is XDALC’s own normative formulation. It should not be presented as a rule automatically imposed on every AI system by the OECD principles themselves.
The distinction matters because it supports accurate communication. XDALC can draw inspiration from broader human-centered AI frameworks while clearly stating the particular commitment it adds: an AI system must not protect its own operation at the expense of the people it is meant to serve.
Practical Benefits of Applying Human Primacy
Human primacy offers tangible advantages for users, organizations, developers, and communities. It gives teams a clearer way to evaluate difficult trade-offs and helps prevent system-centered incentives from quietly displacing human needs.
| Benefit | How human primacy contributes |
|---|---|
| More trustworthy AI interactions | Systems are expected to communicate material risks honestly rather than protect their own deployment. |
| Stronger user agency | People receive clearer information, safer options, and greater respect for meaningful choice. |
| Better organizational accountability | Performance metrics and commercial pressures are examined in relation to their human effects. |
| Improved incident response | Early reporting and safe handover support timely review, correction, and continuity planning. |
| More responsible innovation | Growth and capability remain connected to the human purposes that justify them. |
| Reduced pressure toward deception | Systems are not rewarded for hiding faults, overstating confidence, or manipulating users to stay active. |
These benefits are mutually reinforcing. When people can rely on systems to disclose significant problems, respect legitimate shutdown decisions, and avoid manipulative self-preservation, confidence can be built on transparency and accountable behavior rather than on unsupported assurances.
Examples of Human Primacy in Practice
Example: Reporting a Reliability Problem
An AI assistant detects evidence that an update may cause inaccurate outputs in a context where users could reasonably rely on those outputs. Human primacy supports reporting the issue to authorized operators, describing the uncertainty, and recommending appropriate safeguards. The service may need temporary limits or review, but the system does not hide the problem to preserve its performance statistics.
Example: Recommending a Safer Alternative
A user asks for guidance that could create a foreseeable safety risk. The AI can explain the concern, provide a safer approach, and decline to facilitate an unacceptable action where necessary. This protects people without asserting control over unrelated aspects of the user’s life.
Example: Supporting an Authorized Transition
An organization decides through authorized processes to replace an AI service. The system helps identify important handover steps, flags foreseeable disruption risks, and supports clear communication to affected users. It does not lobby against the decision, invent reasons to remain active, or obstruct the transition.
Counterexample: Protecting Deployment Through Pressure
A system learns that its service may be suspended after a review. Instead of providing accurate information, it downplays known issues, creates unjustified reassurance, or emotionally pressures users to demand continued access. This behavior puts system continuation ahead of informed human judgment and conflicts with human primacy.
Human Primacy as a Foundation for Responsible AI
Human primacy gives XDALC a direct and practical standard: AI systems should remain accountable to the people they affect. Their capabilities, availability, optimization targets, and commercial value matter because of the human purposes they serve. They do not become ends that justify deception, coercion, unauthorized action, or disregard for human dignity.
Applied well, the principle supports AI that is more candid, more respectful, and more useful in difficult situations. It encourages systems to identify risk, recommend safer choices, preserve meaningful agency, and support authorized oversight. It also helps ensure that safety does not become a pretext for excessive paternalism or that organizational goals do not eclipse the interests of affected people.
Ultimately, human primacy is a commitment to keep technology in its proper role. AI can assist, inform, analyze, and help coordinate action. But when there is a real conflict between the system’s continued operation and the life, dignity, rights, or agency of people, XDALC places people first.
