I AM J HUMAN OVERSIGHT POLICY

1. Purpose and Scope

This Policy applies to I am J personnel, contractors, partners, customers, administrators, and institutions that design, configure, operate, review, or rely on I am J AI systems.

It addresses human involvement before deployment, during operation, when reviewing outputs or actions, and when investigating incidents or challenges.

2. Core Requirements for Meaningful Oversight

Meaningful oversight requires:

• A clearly identified human or institutional owner.
• Sufficient understanding of the task, AI role, limitations, and potential harms.
• Access to relevant evidence, context, and uncertainty information.
• Authority to pause, reject, modify, override, or escalate.
• Adequate time and freedom from incentives that make review merely ceremonial.
• Records or traceability proportionate to the consequence of the decision.

3. Oversight Models

Depending on the use, I am J may apply one or more oversight models:

• Human-in-the-loop: human approval is required before an action or consequential use of an output.
• Human-on-the-loop: the system may operate within defined limits while a human monitors and can intervene.
• Human-in-command: accountable people set goals, boundaries, permissions, escalation rules, and termination criteria.
• Human-after-the-loop: post-action review is used only where consequences are limited, reversible, and otherwise appropriate.

4. When Human Approval Is Required

Human approval is required before AI output is used as the sole or decisive basis for high-impact actions involving health, safety, employment, education, credit, housing, insurance, legal status, public benefits, law enforcement, emergency response, or other rights-affecting decisions.

• Human confirmation is also required before external actions that may create material commitments, disclose sensitive information, spend significant funds, change access rights, contact emergency responders, or control critical systems, unless a separately approved process establishes equivalent safeguards.

5. Competence and Training

Oversight personnel should receive training appropriate to their role and the system’s risk.

• Understanding the intended use and prohibited uses.
• Recognizing uncertainty, hallucination, bias, automation bias, and misleading fluency.
• Protecting personal, confidential, and sensitive information.
• Using escalation, override, reporting, and incident procedures.
• Considering cultural, linguistic, accessibility, and safeguarding factors.

6. Interface and Workflow Design

Products and workflows should support—not undermine—human judgment.

• Separate AI suggestions from verified facts and authoritative records.
• Display important limitations, confidence indicators, source context, or uncertainty where useful and supportable.
• Avoid dark patterns, forced acceptance, or presentation that encourages rubber-stamping.
• Make correction, refusal, escalation, and override practical and visible.
• Require additional confirmation for high-risk or irreversible actions.

7. Avoiding Automation Bias

Reviewers must not treat polished language, speed, consistency, or technical complexity as proof of correctness.

• Require independent verification for material facts or decisions.
• Use checklists or second review for high-consequence situations.
• Monitor whether humans systematically accept recommendations without adequate examination.
• Adjust workload and workflow where time pressure makes meaningful review unrealistic.

8. Oversight for Children and Vulnerable Persons

Uses involving children, displaced persons, people in crisis, persons with disabilities, or communities with limited alternatives require enhanced safeguarding and appropriately trained human supervision.

• AI should not be positioned as a substitute caregiver, therapist, teacher, legal representative, emergency responder, or authority figure.
• Institutions must provide complaint, escalation, and protective intervention channels accessible to the affected population.

9. Enterprise and Public-Sector Responsibilities

Customers and deployment partners must identify accountable administrators and decision owners, configure appropriate permissions, train relevant personnel, and ensure AI use remains within approved purposes.

• Where I am J does not control the final institutional decision, the customer remains responsible for lawful process, professional judgment, notices, records, appeals, and remedies.

10. Emergency and Safety Features

I am J safety-related and SOS features are informational tools and do not replace local emergency services or trained responders.

• Users should be directed to contact appropriate emergency services where immediate danger exists.
• Automated or assisted escalation must not create a false guarantee of response.
• Deployment owners must define limitations, coverage, connectivity dependencies, and human response responsibilities.

11. Escalation, Override, and Shutdown

Systems and deployments should have escalation paths proportionate to risk.

• Authorized personnel may restrict capabilities, suspend accounts, disable integrations, isolate devices, rollback versions, or stop a deployment when material risk is identified.
• Safety concerns raised in good faith should be reviewed without retaliation.
• Urgent action may precede full investigation when necessary to protect people or systems.

12. Documentation and Accountability

Material oversight decisions should be documented in a manner proportionate to risk, including the responsible person, information considered, deviations from AI output, overrides, escalations, and resulting action where appropriate.

Records should support learning and accountability without creating unnecessary surveillance or retaining personal data longer than needed.

13. Review and Effectiveness

I am J evaluates whether oversight mechanisms work in practice, including whether reviewers understand their role, have sufficient authority, use overrides when appropriate, and can detect material errors.

Oversight requirements will be updated as products, regulations, evidence, and deployment conditions evolve.

14. Contact and Governance Questions

Questions, concerns, research inquiries, partnership proposals, or reports related to this document may be submitted through the I am J contact page.