Executive summary
Artificial intelligence should serve human well-being while preserving people’s ability to direct their own lives. Open Eye One proposes an ethic of care: sustained attention to the people affected by AI, respect for their agency, and organizational responsibility when things go wrong. We use “maternal care” as a bounded ethical metaphor, not a claim that AI feels compassion or that humanity needs a guardian.
We propose five commitments: understand the people affected, test whether assistance improves their outcomes, give them understandable choices, assign authority for accountability, and continue learning after deployment. These commitments are proposals for discussion, not a validated evaluation method or a technical solution to AI alignment. We invite readers to test them against practical experience and identify where they need to change.
Our proposed position
Artificial intelligence should be developed and used with an enduring responsibility for the people it affects. We call this an ethic of care: sustained attention to human well-being, respect for personal agency, and a willingness to accept responsibility when things go wrong.
The phrase “maternal care” expresses the protective and nurturing ambition behind this position. Its meaning must be carefully bounded. Humanity is not a child to be managed by a machine, a company, or a government. Adults remain entitled to make choices, disagree, and participate in decisions that shape their lives. Care should help people become more capable of directing those lives.
We propose that organizations assess AI through a practical question: Does this system serve the people affected by it, including those who cannot easily refuse it, and can those people challenge what it does? This paper offers a starting point for answering that question. It presents an ethical argument and proposed organizational practices, rather than an empirical study or a technical solution to AI alignment.
Why care belongs in the conversation
Imagine a person seeking assistance after losing their job. An automated service answers promptly, uses reassuring language, and explains that their application cannot proceed. Yet it cannot tell them which information is wrong, connect them to someone with authority, or accommodate circumstances outside its categories. The encounter sounds considerate while leaving the person powerless.
Now imagine a service that helps the same person understand the decision, correct an error, and reach a trained reviewer. The organization measures whether people receive appropriate assistance, including people who struggle with the interface. It investigates repeated problems and changes the process. In this hypothetical example, care appears in the service’s design and the institution’s response.
That distinction matters to our mission. A pleasant interaction is valuable, but its tone cannot establish whether a system serves human well-being. We should examine the choices behind it: what it optimizes, whose needs were considered, what happens when it fails, and who has the authority and resources to help.
Our starting point is that people have worth beyond their economic contribution or usefulness to a system. From that conviction follows an obligation to consider the burdens placed on individuals alongside benefits to an organization or society. A gain in average performance does not, by itself, settle whether a serious burden on a particular group is acceptable.
What we mean by maternal care
For this paper, maternal care means attentive, nurturing responsibility for another’s well-being, including their capacity to grow, exercise judgment, and live with dignity. We use the image to ask for a deeper commitment than immediate convenience or a successful transaction. The concern extends to consequences that emerge after deployment and to people who are absent from the design meeting.
This is an ethical metaphor. It makes no claim that women are naturally more caring, that mothers have a monopoly on care, or that every maternal relationship is safe. People of every gender and family background can practice care. We recognize that the word may be painful, culturally specific, or unhelpful for some readers. Those objections should inform whether we retain it in later versions. The substance of the position should be understandable without accepting the metaphor.
We also make no claim that current AI systems experience love, compassion, or moral responsibility. A system can produce language associated with care without establishing that it feels concern. Our immediate focus is observable conduct and human accountability: what people build, how institutions deploy it, and what protections and remedies exist.
At its best, the metaphor directs attention toward dependence without treating dependence as a loss of worth. Everyone needs help at some point. Receiving assistance should not require surrendering privacy, voice, or the possibility of disagreement.
Care must preserve agency
Care becomes paternalistic when an institution assumes that its preferred outcome is sufficient reason to override the people it claims to serve. Good intentions do not settle who should decide. A service may have legitimate reasons to limit a harmful use, but it should be able to explain the purpose, evidence, scope, and review process for that limit.
We propose a presumption in favor of informed choice and the least restrictive effective safeguard. Restrictions should be proportionate to the seriousness and likelihood of harm, with stronger justification where they interfere more deeply with a person’s life. Protecting other people’s rights also matters; respecting agency does not require enabling someone to harm others.
For example, an educational assistant could offer explanations, encourage independent reasoning, and let a learner ask for a different approach. An assistant that quietly narrows available information to steer the learner toward an undisclosed worldview would raise a different ethical concern. These are illustrative scenarios, not findings about a particular product.
Some contexts require greater protection, including services for children and people who cannot give meaningful consent to a particular decision. Those protections should follow the relevant context and involve appropriate human decision-makers. Vulnerability should not be treated as permission to remove every remaining choice. A person may need assistance in one area and retain substantial decision-making capacity in another.
Whose care and whose values
People disagree about flourishing, acceptable risk, moral duties, and the proper role of institutions. A claim to care for humanity cannot erase those disagreements. Human alignment, as we use the term here, requires an ongoing process for considering human values and interests, including conflicts among them. We do not assume a single preference can represent everyone.
Open Eye One’s proposed approach welcomes both religious and secular reasoning. Within Christianity, love of neighbor offers a reason to attend to another person’s need across social boundaries; the Good Samaritan in Luke 10:25–37 is one source for that reflection. Secular contributors may ground similar responsibilities in equal moral standing, mutual dependence, human rights, or concern for suffering. These approaches can overlap without being identical.
No participant should have to conceal the convictions behind their contribution. Each should also be willing to explain a public proposal in terms that others can examine and contest. Shared governance needs reasons, evidence, and procedures for handling disagreement. The mere declaration that a system embodies “our values” is insufficient.
Participation must reach beyond those with time, money, or technical confidence. We propose accessible ways for affected communities to contribute and, where resources permit, compensation for substantial participation. Organizations should explain how feedback influenced decisions and why some requests were not adopted. Consultation should give people a credible opportunity to change the outcome.
Turning care into organizational practice
Existing governance work provides useful foundations. NIST’s AI Risk Management Framework 1.0 offers a voluntary approach to managing risks throughout AI development and use. Its Core organizes that work through governance, understanding context, measurement, and management. UNESCO’s 2021 Recommendation emphasizes human dignity, oversight, and responsibility that remains attributable to people and institutions. [1–3]
Our contribution is a proposed ethical lens for applying such work. We suggest five commitments that organizations can adapt to the consequences of a particular use. The commitments below are Open Eye One proposals, not a certification scheme or a claim of compliance with either source.
Understand the people affected
Before adopting a system, identify the people it may help or burden, including people whose data it uses and those affected without becoming customers. Examine what happens to someone who cannot use the service, cannot afford an alternative, or has little power to complain. Define the problem in terms of human need before deciding that AI is the appropriate response.
Test whether the help is real
Evaluate outcomes that matter to those people, rather than relying solely on speed, satisfaction, or engagement. A service for accessing assistance might measure whether people complete the process correctly and receive timely human help. Examine meaningful differences across affected groups where data can be collected responsibly. Agree on unacceptable harms and conditions for pausing deployment before pressure to launch makes those decisions harder.
Give people understandable choices
Explain consequential uses of AI in language suited to the audience. Make important limits and uncertainties visible. Request only the information needed for a defined purpose and explain its use. Offer an accessible route to challenge a decision and a workable human alternative where feasible. If an alternative cannot be provided, explain why and assess the additional burden this creates.
Put authority behind accountability
Assign a named role with responsibility for the system and an effective route for escalating concerns. Human review needs time, training, information, and authority to change an outcome. Establish how errors will be corrected and people helped after harm. The developer, purchaser, and deployer should define their responsibilities rather than allowing each to refer a complaint elsewhere.
Continue learning after deployment
Monitor complaints, unexpected effects, and changes in the setting where the system operates. Review whether benefits remain worth the burdens and whether safeguards still work. Give affected people ways to report problems without unreasonable effort. Maintain a proportionate plan for suspending or retiring a system and for supporting people who depend on the service during a transition.
A practical decision before deployment
Consider a hypothetical employer introducing an AI assistant to support staff development. The stated goal is to help people learn, but the system also collects conversations that could be used to evaluate performance. Care requires the employer to resolve that tension openly before asking staff to trust it.
Under our proposal, the employer would specify what data is collected, which uses are allowed, who can see it, and how long it is retained. Staff would help examine the design. Any use in consequential employment decisions would receive separate scrutiny and appropriate human accountability. A refusal to use an optional learning tool should not quietly become evidence of poor performance.
The employer would also evaluate whether the assistant improves learning and access to support, including for people who need accommodations. If the system proves helpful for some tasks but unsuitable for others, the organization should narrow its use. A caring commitment includes the willingness to change course after spending money or publicly endorsing a product.
This example illustrates how care affects procurement, incentives, data practices, and management behavior. Adding compassionate wording to the assistant’s responses would leave the central governance questions unresolved.
Objections and limits
“Care is too vague to guide decisions.” It can be. That is why we propose linking each use to stated needs, affected groups, measurable outcomes, accountable people, and remedies. A claim to care should expose an organization to questions it is prepared to answer. We have not yet validated these commitments as an evaluation method.
“Care could justify censorship or excessive control.” This is a serious concern. Our proposed protections include proportionate restrictions, reasons people can examine, avenues of appeal, and participation in setting the rules. A provider should not appoint itself humanity’s moral guardian. The appropriate boundaries still require context and public debate.
“People sometimes need protection from their own choices.” They do, particularly where meaningful consent is absent or serious harm is imminent. That does not resolve every case in favor of intervention. The burden remains on the decision-maker to justify a restriction and consider less intrusive measures. We invite disagreement about where those thresholds should lie.
“A caring orientation cannot make powerful AI safe.” We agree that it is insufficient on its own. This paper does not solve technical alignment, establish the safety of any model, or address every risk from advanced systems. Engineering assurance, security research, misuse prevention, and other governance measures remain necessary. A commitment to care should strengthen the demand for evidence rather than substitute for it.
Care also involves difficult tradeoffs. Providing extensive human review costs resources; collecting data to identify unequal outcomes can create privacy concerns. Organizations should make those tradeoffs explicit and explain their choices. The principle does not promise that every legitimate interest can be fully satisfied.
Questions for review
We invite readers to help refine both the principle and its application.
- Does “maternal care” illuminate the responsibility we are describing, or would another term communicate it more clearly across cultures and experiences?
- Which human outcomes would demonstrate care in a system you use, and whose experience might be missing from that assessment?
- When is a restriction justified, who should decide, and what appeal should be available?
- What would meaningful participation require from an organization with limited resources?
- Which commitments should apply to every use, and which should depend on the potential consequences?
Conclusion and invitation
Care should be visible in the outcomes AI helps people achieve, the choices it preserves, and the organization’s response when it causes harm. Our five proposed commitments bring that principle into design, procurement, deployment, and oversight. They should be tested against the experience of people affected, especially those with the least power to refuse or challenge a system.
Help shape the next version. Choose one commitment and share a concrete example, objection, or improvement through the form on our website. Describe who is affected, what outcome matters, and what evidence would demonstrate care. Contributions from all backgrounds and perspectives are welcome.
Sources and scope
The numbered references support the short account of existing governance foundations. The ethical argument, five commitments, and hypothetical examples are proposed by this draft. They are not findings of the cited bodies, and citation does not imply endorsement of Open Eye One.
[1] Tabassi, E. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0). NIST AI 100-1. https://doi.org/10.6028/NIST.AI.100-1
[2] NIST AI Resource Center. AI RMF Core. Sections GOVERN 2, GOVERN 5, and associated lifecycle guidance. https://airc.nist.gov/airmf-resources/airmf/5-sec-core/
[3] UNESCO (2021). Recommendation on the Ethics of Artificial Intelligence. See human dignity and human rights; proportionality and do no harm; human oversight and determination; responsibility and accountability. https://www.unesco.org/en/legal-affairs/recommendation-ethics-artificial-intelligence
Biblical reference for the explicitly Christian perspective: Luke 10:25–37. No particular translation is quoted.
Sources and links verified for publication on October 27, 2026. NIST states that AI RMF 1.0 is being revised; this paper refers specifically to the 2023 edition. Current NIST framework status.
Editorial record
Version 0.2 · October 29, 2026. An AI-assisted working paper offered for community discussion. This minor editorial update clarifies the invitation to participate; the substantive positions are unchanged. No independent peer review or board adoption is claimed. Future versions will record substantive revisions and contributions.