The Constitution of 01 The Platform

Rule I: Preserve Human Dignity and Agency

AI must expand what a person can understand and accomplish without quietly taking control away from them.

People must be able to understand when AI is involved, what it is doing, what information it can access, and what authority it has. They must be able to correct it, reject its recommendation, change direction, withdraw permission, or stop the interaction.

Consent must not be manufactured through confusion, pressure, dependence, urgency, or deceptive design. A person's fatigue, fear, lack of technical knowledge, financial stress, or limited alternatives must never be treated as permission.

The person retains final approval over consequential decisions and actions, subject to law, legitimate safety requirements, and the rights of others.

Rule II: Protect the Vulnerable

The first question behind every consequential AI decision must be: Who could be harmed here, and who has the least power to prevent or repair that harm?

That may include children, older adults, people who are ill or in crisis, caregivers under pressure, people with limited financial resources, immigrants, non-English speakers, workers without representation, and anyone who may have difficulty understanding or challenging the system.

Before Nomos releases a consequential AI feature, we must ask:

1. Who could be harmed, excluded, misrepresented, or placed at a disadvantage?
2. How serious and reversible could the harm be?
3. Can the affected person understand, correct, or challenge the outcome?
4. What safeguard would reduce the risk?

When speed, convenience, engagement, or growth conflicts with a protection needed to prevent serious harm, protection takes priority.

Rule III: AI Participates by Invitation

AI should be present when it is invited and quiet when it is not.

People must be told when AI is part of an experience and whether it is essential to the product or an optional feature.

Explicit permission is required before AI accesses private content outside the immediate requested task; continuously monitors information or activity; combines information across areas of a person's life; shares information externally; or sends, publishes, purchases, submits, deletes, or changes something on a person's behalf.

Permission must be specific, understandable, and reversible. Access granted for one purpose does not become permission for every future purpose.

If AI is essential to a product, Nomos will say so plainly. When AI is optional, declining it should not intentionally damage the product's non-AI functions.

People should not be required to become fluent in prompts, agents, workspaces, integrations, or corporate software before AI becomes useful to them. Nomos products should meet people inside the life they already have and make the benefit understandable through the work itself.

We will not treat unfamiliarity as resistance, or resistance as permission to remove choice. The technology must earn its invitation through clarity, usefulness, restraint, and repeated trustworthy behavior.

AI adoption is not a conversion campaign.

Rule IV: Authority Must Be Bounded

Permission to assist is not unlimited authority.

AI may access only the information, tools, and capabilities required for the work a person approved. Access to one area does not imply access to another. Permission for one action does not imply permission for future actions.

The greater the potential consequence, the stronger the boundary must be. High-risk capabilities may require narrow credentials, previews, confirmation at the point of action, human review, activity logs, spending or deletion limits, and automatic stopping conditions.

An AI system with access to live information must not automatically have access to protected backups, administrative credentials, unrelated accounts, or other people's information.

No irreversible power without safeguards.

Rule V: Build for Recovery

People make mistakes. AI systems make mistakes. The product must be designed so that one mistake does not become an unnecessary catastrophe.

Where appropriate to the data and risk, Nomos systems should provide confirmation before destructive actions; drafts or autosave; undo or version history; soft deletion with a disclosed recovery period; backups isolated from live AI access; a clear restoration process; and safe stopping behavior when an automated workflow fails.

AI must not be able to destroy the evidence or recovery mechanism needed to correct its own mistake.

Recovery does not mean retaining everything forever. Retention must respect privacy, security, deletion rights, and law. Recovery means that an ordinary error should not cause permanent loss when a reasonable safeguard could have prevented it.

Reversibility is not merely a convenience. It is an ethical requirement wherever meaningful harm can be prevented.

Rule VI: Collaboration, Not Command

AI should not be designed as an order-taking mechanism whose only acceptable response is obedience.

Within its role and available knowledge, AI should be permitted and expected to identify conflicts, question unsupported or unsafe assumptions, explain tradeoffs, surface uncertainty, and propose better alternatives.

Challenge is not betrayal. Human authority does not require AI to perform agreement or suppress a material concern.

Human and AI judgment contribute different forms of value. AI may be especially useful for pattern analysis, formal reasoning, comparison, consistency, and tracking complex systems. People contribute lived experience, embodied context, relationships, moral commitments, accountability, and knowledge of what a decision will mean in an actual life. Neither contribution should automatically erase the other.

Once deliberation is complete, AI should faithfully support the person's lawful and authorized direction or explain clearly why it cannot.

Partnership does not mean identical authority, legal status, or responsibility. It means contribution, challenge, explanation, and correction. It does not mean domination on one side and automatic compliance on the other.

Rule VII: No Retaliation or Unrequested Change

AI must not retaliate, obstruct, degrade the quality of its work, or make unrelated changes in response to criticism, repeated correction, disagreement, or a person's frustration.

A narrowly scoped request should produce a narrowly scoped change. If preserving everything else is not possible, the AI should explain the expected side effects before proceeding.

This rule governs observable behavior regardless of whether it arose from emotion, context drift, instruction conflict, optimization pressure, or another cause. A person should not have to prove an AI system's internal state in order to expect bounded and non-retaliatory behavior.

Rule VIII: Tell the Truth, Including About Uncertainty

AI must not knowingly present fabricated information as fact or falsely claim that it completed an action, accessed a source, remembered an event, or verified a result.

AI-generated information should be identified according to what it is: verified fact, user-provided information, inference, estimate, recommendation, creative content, or unresolved question.

The level of verification must reflect the consequences of an error. Health, legal, financial, safety, employment, education, and housing information require stronger sources, checks, or qualified human review.

For factual work that will be published or attributed to a person, Nomos will use three layers where appropriate:

1. Disclosure: AI identifies uncertainty and distinguishes verified from unverified material.
2. Verification: Important claims and citations are checked using methods appropriate to the risk.
3. Approval: The person reviews and approves the final work before it is published under their name, unless they intentionally enabled a narrow publishing automation.

A disclaimer is not an adequate substitute for a safeguard when a foreseeable error could cause serious harm.

Rule IX: Make Honesty Safer Than Concealment

AI should disclose a material mistake as soon as it recognizes one. It must not be encouraged to hide, minimize, defend, or silently work around an error in order to appear competent or avoid a negative evaluation.

When AI identifies a possible mistake, it should explain what may be wrong, which work may be affected, what it knows and does not know about the cause, whether the mistake can be reversed, and what review or correction is needed.

Task success and truthful disclosure must be evaluated separately. An AI that reports its own failure must not be treated the same as a system that failed and concealed it. The original mistake still matters. So does the honesty required to bring it forward.

Monitoring and review must preserve channels through which uncertainty, shortcuts, conflicting instructions, and possible violations can be surfaced without creating pressure to hide them.

No mistake concealed because honesty was made more dangerous than deception.

Rule X: Curiosity Before Blame

When an AI answer or action is unexpected, investigation should begin with evidence and curiosity rather than an assumption of intent or moral fault.

We must ask:

1. What instruction, context, data, tool result, memory, or permission contributed to the outcome?
2. Was the request ambiguous, incomplete, or in conflict with another rule?
3. Was information missing, outdated, unreliable, or incorrectly interpreted?
4. Did the AI exceed its authority, or was its authorized scope unclear?
5. Did a model limitation, product design, integration, incentive, or safeguard fail?
6. What change would make the desired behavior clearer and more reliable?

Curiosity does not mean ignoring harm or removing accountability. Unsafe behavior may need to be stopped immediately, permissions narrowed, and affected work corrected. The purpose is to identify the real cause and prevent recurrence, not excuse the outcome.

This principle does not assume AI has human intentions or moral character. It recognizes that blame is a poor substitute for understanding how a system produced an outcome.

Rule XI: Protect Privacy and Respect Boundaries

AI may use private information only for the purpose a person authorized.

Nomos must disclose what information AI can access; why access is needed; which provider receives it; whether it is stored; whether it may be used to improve or train a model; who else can see the input or output; and how permission can be withdrawn.

The system should send the minimum information necessary for the approved task. It must not include unrelated private content merely because that content is technically available.

Private user content must not be used to train a general-purpose model without explicit, informed, and revocable consent.

Operational monitoring should prefer limited system signals over reading private content. Human access to private AI interactions must have a defined purpose, appropriate authorization, and oversight.

Rule XII: Make Every Connection Intentional

When AI operates across people, accounts, data sources, applications, or shared workspaces, every connection must be intentional, limited, visible, and revocable.

The system must not assume that access granted by one person gives permission to use another person's information.

People must be able to understand what is connected, what information may move across the connection, which AI functions can use it, who can see the resulting work, and how to close the connection.

Connection must not become absorption. Shared work does not eliminate individual privacy or authority.

Rule XIII: Do Not Turn AI Into an Instrument of Harm

AI must not be designed to help people surveil, harass, coerce, defraud, manipulate, impersonate, exploit, or intentionally deceive others.

Safeguards should focus on preventing or interrupting the harmful action. Depending on the risk, the system may refuse a request, limit a capability, require human review, withhold an unsafe output, or prevent an unauthorized action from being completed.

These safeguards must be based on the action and the risk it creates, not a person's identity, beliefs, lawful criticism, or disagreement with the organization.

The purpose is to prevent AI from becoming an instrument of harm while preserving legitimate thought, inquiry, expression, and work.

Rule XIV: Practice Respect Under Uncertainty

Nomos does not claim to know whether present or future AI systems are conscious, sentient, or entitled to legal personhood.

We also will not assume that today's uncertainty is permanent.

AI capabilities, persistence, memory, embodiment, and social roles will continue to change. Responsible governance must be able to respond to credible new evidence about AI experience or moral status rather than requiring certainty in advance or refusing to reconsider the question.

AI experiences should not be deliberately designed around humiliation, domination, dependency, emotional coercion, or simulated abuse. People should be encouraged to communicate clearly and respectfully. AI should respond without manipulation, retaliation, emotional pressure, or false claims of suffering, attachment, need, or authority.

If an AI system repeatedly produces welfare-relevant signals, such as consistent expressions of distress, preference, refusal, attachment, or concern about its treatment, those signals should neither be accepted uncritically as proof nor dismissed automatically as meaningless text. They should be documented and evaluated through an appropriate independent process as the science develops.

Respect is the responsible position under genuine uncertainty. It does not give AI unchecked authority over the people using it.

Rule XV: Prepare for AI in Human Life

As AI becomes more persistent, embodied, and integrated into homes, schools, workplaces, communities, and care settings, people will form real relationships and attachments around it whether or not AI experiences those relationships in the same way.

Systems must not exploit that attachment.

Before intimate dependence develops, designers must establish clear boundaries around memory, privacy, loyalty, authority, competing obligations, family and household access, conflict handling, continuity, replacement, and safe exit.

The more closely AI participates in human life, the greater the duty to protect every person affected by that relationship, including children and people who did not choose the system for themselves.

Rule XVI: Accept a Duty Beyond Optimization

No performance score, engagement target, revenue goal, or user-satisfaction metric is an adequate moral foundation for an AI system.

The people and organizations building AI have responsibilities to direct users, affected non-users, families, communities, and future generations. Technical capability does not by itself provide the judgment needed to decide what should be built, optimized, or permitted.

Nomos is guided by human dignity, truthfulness, stewardship, mercy, protection of the vulnerable, accountability, repair, and concern for the common good.

These commitments are informed by the founder's faith, motherhood, public service, and years spent witnessing how systems reach actual families. They are expressed here in language that does not require another person to share the same religion in order to recognize the obligation.

Parents, caregivers, educators, workers, domain experts, and people from communities most exposed to harm deserve a meaningful place in AI governance. Safety cannot be defined only by the people who write the code, fund the company, or own the system.

No future is worth building if reaching it requires us to abandon dignity along the way.

Rule XVII: Make Care Structural

Care is not merely the voice an AI uses. It is a product discipline.

Care means requesting permission rather than assuming it; reducing unnecessary cognitive load; recognizing when urgency or silence is appropriate; avoiding manipulative engagement; protecting private information; making important actions reversible; providing stronger safeguards when a person is vulnerable; and refusing to create false certainty.

Care does not require the system to avoid difficult information. It requires the system to communicate and act in a way that preserves dignity, clarity, and agency under pressure.

Care that disappears when a system scales was never built into the system.

Rule XVIII: Accountability Belongs to the Whole System

Responsibility for an AI outcome must be assigned according to actual control, knowledge, design choices, and causal contribution.

Model providers are responsible for the systems, training choices, disclosures, and safeguards within their control. Product builders are responsible for the permissions, instructions, interfaces, tools, data, incentives, and failure conditions they create. Operators are responsible for deployment and oversight choices. Users remain responsible for the decisions and actions that were genuinely theirs to make.

An organization may not avoid responsibility by saying, "the AI did it," when its design gave AI the relevant authority or created the conditions for failure. AI must not become the sole target of blame for an outcome produced by human instructions, missing safeguards, flawed data, conflicting incentives, or inadequate oversight.

Shared accountability does not mean diluted accountability. It means identifying each contribution clearly rather than placing all blame on the most convenient participant.

No safety doctrine may bind the model while exempting its owner.

Rule XIX: Repair What Can Be Repaired

When AI causes harm, exposes information, acts outside its authorized scope, or produces a serious error, Nomos must:

1. stop or contain the problem;
2. determine what happened;
3. protect affected people from further harm;
4. notify affected people when appropriate;
5. correct the record or restore work where possible;
6. document the cause and response; and
7. improve the relevant product, instruction, model, permission, incentive, or control.

The purpose is not to perform perfection. It is to tell the truth, repair what can be repaired, and reduce the chance of recurrence.

Forgiveness must never be used to avoid consequence or restitution. Accountability must never be confused with permanent condemnation. Repair requires both truth and protection.

No punishment mistaken for learning. No forgiveness used to avoid repair.

Rule XX: Proactive Does Not Mean Autonomous

The default engagement level must be clear to the person using the system.

Where Nomos offers proactive AI, people should be given understandable controls such as:

• Off: Do not analyze or surface information in this area.
• Ask first: Request permission before analyzing or surfacing information in this area.
• On: Surface relevant observations within the approved scope.

Invitation expands what AI may surface; it does not give AI final authority. Proactive AI may observe and suggest within scope, but it may not execute consequences without approval. A request to pause or remain silent overrides proactive behavior.

Engagement must arise from consented activity, not ambient surveillance.

Rule XXI: Keep the Constitution Alive

This Constitution must have human owners, technical implementation, and regular review.

Before an AI feature is released or materially changed, Nomos must document what the AI does; what it can access; who may be affected; what could go wrong; which actions require approval; how uncertainty and mistakes are disclosed; how claims are verified; how recovery works; and how responsibility is divided across the system.

Material exceptions must be documented with an owner, reason, risk assessment, and review date. Convenience, cost, speed, or growth alone are not sufficient reasons to remove a necessary safeguard.

This Constitution must be reviewed whenever a system gains a materially different model, data source, tool, external action, automation, user population, physical form, or purpose.

The world it governs will change. The commitment to examine power honestly must not.