01
Findings remain independent
Money, access, sponsorship, and client relationships do not purchase a favorable conclusion.
Governance
Trust requires rules that hold under pressure.
Fide AI publishes the rules that govern how we are funded, how we handle conflicts, how we report findings, and where decision-making authority remains. These commitments make our work easier to inspect before anyone is asked to trust it.
02
Claims remain bounded
Results travel with their system, conditions, assumptions, uncertainty, and stated limits.
03
Authority remains human
Fide AI provides evidence and guidance. Responsible people and institutions retain the decision.
No pay-for-rank
Sponsorship, participation, advisory fees, discounted work, pro bono support, or commercial relationships cannot purchase favorable findings, recommendations, certification, release timing, launch approval, buying support, or marketing claims. Fees and discounts buy or subsidize staff time and analysis, not outcomes.
Related-entity controls
Any related client or participant, including organizations with financial relationships to Fide AI personnel, receives enhanced review, disclosure, recusal from scoring or advisory decisions, and non-conflicted signoff before findings are finalized.
Claims humility
Scores and advisory recommendations are scoped judgments under named assumptions, not theological authority, pastoral authority, universal product approval, or certification of safety. Public reports must state the condition name, model version, and calibration status when applicable.
Assumption and worldview disclosure
Research and evaluation materials should state the material assumptions, value judgments, disciplinary perspectives, domain standards, and worldviews that shape question selection, criteria, interpretation, or recommendations. Disclosure makes the reasoning inspectable; it does not substitute for evidence, validation, or stated limits.
Monitoring proportionality
Agent monitoring should collect the minimum operational evidence needed for a defined safety, evaluation, or accountability purpose. Access to traces, employee activity, user interactions, and confidential context must be limited, documented, and governed rather than treated as unrestricted surveillance.
Public and private accountability
Private client work may protect confidential materials, but methods, limits, correction processes, conflict controls, and public claims must stay understandable. Fide AI does not hide scoring rules, review rules, or reviewer conflicts when public claims are made.
Deployment authority
Fide AI provides research, evaluation, design guidance, and implementation patterns. Churches, ministries, schools, parents, institutions, and builders retain responsibility for whether, where, and how AI systems are deployed.
Sponsor and client non-interference
Sponsors, clients, and contributors may not direct findings, recommendations, publication timing, marketing claims, or suppression of unfavorable results. Sponsorship relationships are disclosed in public materials.
Evaluator scrutiny
When Fide AI publishes evaluation claims, the work should be legible to outside evaluators: access conditions, methodology, conflict handling, release limits, and claim boundaries should be stated plainly enough for independent readers to assess credibility.
Policies in practice
Useful work without compromised judgment.
Fide AI may offer paid evaluations, advisory services, sponsorship opportunities, pilots, product prototypes, and implementation support. These activities are permitted when they preserve claims limits, conflict controls, and the authority of the people and institutions making deployment decisions.
Read the full independence policy →Product Lab Boundaries
Prototypes and sample designs are research-backed product work unless a separate service agreement says otherwise. If Fide AI evaluates a system it helped design, that involvement must be disclosed and reviewed with stronger independence controls.
Claims and Certification
Evaluation results are tied to a specific system, version, setup, and use case. Product-lab participation, advisory support, sponsorship, or benchmark participation does not equal endorsement, certification, or approval unless a separate public Fide AI report explicitly supports that claim.
Conflicts and Related Entities
Related organizations, sponsors, paid clients, sponsored pilots, and product partners can participate only under disclosure, recusal, non-interference, and claims-limit rules.
Human Care and Authority
Fide AI does not promote AI as clergy, pastor, counselor, spiritual director, parent, or substitute for responsible human authority. Systems should be evaluated for boundaries, escalation, and human handoff.
Parent and Minor Safety
Child-facing and family-facing AI requires evaluation for age-appropriate disclosure, tutoring boundaries, companionship and dependency risks, parental visibility, privacy, and escalation for self-harm, abuse, grooming, or unsafe dependence.
Data and Evaluation Records
Paid and unpaid work should define what data is collected, what records are kept, when results can be used for research, and what privacy-protecting summaries may be published.
Operational Trace Governance
Work involving agents or production monitoring should define which operational traces are necessary, who can access them, how long they are retained, how sensitive content is protected, and when data may be used for secondary research. Trace access does not imply access to private chain-of-thought.
External Standards Fit
Fide AI may participate in broader AI evaluation communities when doing so strengthens public-interest evaluation practice without weakening its responsibilities to faith-facing work, claims limits, or high-trust institutions.
Institutional boundaries
Clarity about what our work does not confer.
A vendor ranking marketplace
A pay-for-rank certification shop
A reseller or sales agency for any product vendor
A theological or pastoral authority
A replacement for church, school, or counseling
The marketing arm of any product company