Ethical AI Policy.

Document Control

1. Purpose and Public Statement

House of Wisdom builds Decision Infrastructure: the governed layer through which complex organisations turn information into trusted decisions. Artificial intelligence is central to how we do this — but always as a tool in service of people, never as a replacement for human judgement, authority or accountability.

Our position is simple. AI should enrich human knowledge, sharpen human reasoning and improve human decision-making — under meaningful human control, with full traceability, and for the benefit of the organisations and people we serve. Intelligence is generated by machines; decisions are owned by people.

This policy sets out the principles, controls and accountabilities that govern every use of AI within House of Wisdom and across our platform — including the Wise Owl intelligence layer and its specialist agents. It is written to be auditable and is aligned with the recognised international frameworks for trustworthy and responsible AI.

2. Scope and Applicability

This policy applies to all House of Wisdom employees, officers, contractors and partners, and to the full lifecycle of every AI capability we design, deploy, operate or retire — including data sourcing, model selection, agent orchestration, evaluation, deployment, monitoring and decommissioning.

It governs, in particular: the Wise Owl intelligence layer; the Research, Citation, Governance, Risk, Mapping, Solution and Learning agents; the generation of Decision Receipts; and the retention of organisational memory.

Where this policy and applicable law differ, the stricter requirement applies. Nothing in this policy reduces obligations under data-protection law, sector regulation or contractual commitments to customers.

3. Definitions

• AI system — software that, for a given set of objectives, generates outputs such as analysis, options, predictions or recommendations that influence human decisions.
• Human-in-the-loop (HITL) — a design and operating model in which a competent person reviews, can override, and is accountable for the use of AI output before it informs or becomes a decision.
• Wise Owl — House of Wisdom’s intelligence layer, which orchestrates specialist AI agents under governance to produce decision-ready intelligence for human evaluation.
• Decision Receipt — the structured, auditable record attached to a decision, capturing its evidence, sources, contributors, assumptions, risks, rationale and outcome.
• Organisational memory — the retained, reusable record of decisions and their reasoning that allows an organisation to learn and compound capability over time.

4. Our Ethical Principles

Our use of AI is governed by eight principles, aligned with the OECD AI Principles and the trustworthy-AI characteristics of the NIST AI Risk Management Framework.

• Human agency and oversight. People direct, review and remain accountable for AI-assisted work. AI never holds final decision authority.
• Beneficial purpose. We apply AI to enrich human knowledge and improve problem-solving and decision quality — for the organisation and the wider good, not to deskill or displace human judgement.
• Transparency and explainability. Every AI-assisted conclusion is traceable to its evidence and method, and is recorded in a Decision Receipt.
• Accountability. Every decision has a named human owner. Responsibility is never delegated to a machine.
• Fairness. We actively identify and manage harmful bias in data, models and outputs.
• Privacy and data stewardship. We protect personal and confidential information and process it lawfully and proportionately.
• Safety, security and robustness. Our systems are tested, monitored, secured and resilient across their lifecycle.
• Contestability and remedy. AI output can be questioned, overridden and corrected, and there is a clear route to challenge and remediate.

5. Human-in-the-Loop and Human Oversight

Human oversight is the foundation of Decision Infrastructure. It is a design requirement, not an afterthought.

Consistent with Article 14 of the EU AI Act and the UNESCO Recommendation on the Ethics of AI, our systems are built so that competent people can understand, monitor, interpret, override and — where appropriate — decline to use AI output. An AI system can inform a decision; it can never replace ultimate human responsibility.

5.1 Oversight commitments

• AI generates evidence, options and analysis; humans make and own the decision.
• Every governed decision requires explicit human authorisation by a named, competent owner before it takes effect.
• Operators are equipped to understand each system’s capabilities and limitations, to detect anomalies and failures, and to intervene or stop the system.
• Automation-bias safeguards. We design against over-reliance: outputs are presented with their evidence, confidence and limitations so that people engage critically rather than defer automatically.
• The right to disregard. Any operator may override or set aside AI output; doing so is recorded, not penalised.
• No autonomous high-stakes decisions. Decisions with significant impact on people’s rights, safety, livelihoods or wellbeing are never ceded to AI and always require human determination.

The Decision Receipt is our oversight evidence. It records who decided, on what basis, with what evidence and authority — making human oversight demonstrable and auditable after the fact.

6. The Case for Human + AI: Benchmarks and Benefit

We hold that the right unit of performance is not the machine alone, nor the human alone, but the human–AI team — the “centaur”. Our policy is to engineer for that partnership and to measure its benefit.

6.1 What the evidence shows

Independent research consistently finds that well-designed human–AI collaboration outperforms either humans or AI working alone:

• In a large enterprise field experiment, professionals working with AI matched the output quality of two-person teams working without it, and AI-enabled teams produced more top-tier solutions.
• In controlled studies of knowledge work, human–AI teams produced materially more output per person than human-only teams.
• On complex problem-solving benchmarks, human–AI collaboration achieved success rates well above either humans or AI alone.

The conclusion is consistent: AI augments human capability most when a person stays in the loop, interrogates the output, and remains accountable for the result. That is precisely the model Decision Infrastructure operationalises.

6.2 How we measure benefit

We do not deploy AI for novelty. Every capability must demonstrably improve human decision-making against measurable indicators. Our benefit benchmarks include:

• Decision quality — the measured improvement in decision outcomes against a human-only baseline.
• Time-to-decision — the reduction in time from problem identification to a governed decision.
• Knowledge retention — the share of decisions, reasoning and lessons captured as reusable organisational memory.
• Traceability coverage — the proportion of decisions supported by a complete, auditable Decision Receipt.
• Human engagement — the breadth of expert participation and the rate at which people review, challenge or override AI output.
• Capability building — evidence that people and the organisation become more capable over time, not more dependent.

Independent evaluation of the House of Wisdom platform across healthcare, defence and government settings recorded strong results on these dimensions, including decision-quality uplift, faster time-to-decision and increased institutional knowledge retention versus baseline.

7. AI Lifecycle Governance and Controls

We operate an AI management system aligned with ISO/IEC 42001 and structured around the four functions of the NIST AI Risk Management Framework — Govern, Map, Measure and Manage.

7.1 Govern

• An AI Governance Board owns this policy, approves higher-risk use cases and reviews incidents.
• Clear roles, responsibilities and competency requirements are defined for everyone involved in the AI lifecycle.

7.2 Map

• Every AI use case is assessed for context, purpose, affected people and potential impact before development or deployment.
• An AI system impact assessment is completed and recorded for each capability, and revisited on material change.

7.3 Measure

• Systems are tested for validity, reliability, safety, security, bias and explainability before release and on an ongoing basis.
• We track trustworthiness metrics and benefit benchmarks, and document system functionality and limitations.

7.4 Manage

• Risks are prioritised and mitigated; residual risk is owned and accepted by a named authority.
• Issues, anomalies and incidents trigger documented response, intervention and, where needed, rollback or shutdown.

8. Accountability and Roles

• AI Governance Board — policy ownership, approvals for higher-risk uses, incident oversight, annual review.
• Decision owners — named individuals accountable for decisions informed by AI.
• System owners — accountable for the safe, compliant operation of each AI capability.
• All personnel — responsible for using AI in line with this policy and for raising concerns.

9. Transparency, Provenance and Decision Receipts

We make AI-assisted reasoning visible and reviewable. Where AI contributes to analysis, that contribution is disclosed. Sources and evidence are cited and validated; provenance is preserved.

• Citation and provenance. The Citation agent validates sources so conclusions can be traced to evidence.
• Decision Receipts. Every governed decision is recorded with its evidence, sources, contributors, assumptions, risks, rationale and outcome — making decisions transparent, auditable and defensible.
• Disclosure. Users are informed when they are interacting with, or relying on, AI-generated material.

10. Data Governance and Privacy

• Personal and confidential data is processed lawfully, fairly, transparently and only as necessary for a specified purpose.
• We apply data minimisation, access control, retention limits and security appropriate to the sensitivity of the data.
• Customer and organisational knowledge remains the customer’s; it is not used beyond agreed purposes.
• Data quality is actively managed, because decision quality depends on input quality.

11. Safety, Security and Robustness

• AI systems are tested for accuracy and resilience, and monitored for drift, degradation and unexpected behaviour.
• Security controls protect against misuse, manipulation (including adversarial and prompt-based attacks) and unauthorised access.
• Fallback and human-intervention paths exist so operation can continue safely if a system fails or is withdrawn.

12. Fairness and Bias Management

• We assess data and models for harmful bias and discriminatory impact across affected groups.
• We challenge assumptions and narrow datasets; the Risk agent surfaces uncertainty and the Governance agent applies policy controls.
• Identified bias is documented, mitigated and re-tested; unresolved material bias blocks deployment.

13. Prohibited and High-Risk Uses

House of Wisdom will not develop or knowingly enable AI that:

• makes consequential decisions about people without meaningful human oversight;
• is used for unlawful surveillance, social scoring, manipulation or deception;
• cedes life-and-death or other high-stakes determinations to a machine;
• operates without traceability, accountability or the ability to be overridden; or
• is deployed for a purpose, or in a context, for which it has not been assessed and approved.

High-risk use cases require explicit AI Governance Board approval, a completed impact assessment and enhanced oversight before deployment.

14. Monitoring, Audit and Continual Improvement

• AI capabilities are monitored in operation, with post-deployment review of performance, benefit and incidents.
• This policy and our controls are audited and reviewed at least annually, and upon material change to law, standards or our systems.
• Lessons learned are captured in organisational memory and fed back into design — our own Action Learning, applied to ourselves.
• A clear route exists for staff, customers and the public to raise concerns or contest AI-influenced outcomes.

15. Standards and Control Mapping

This policy is designed to support enterprise assurance and procurement. The table below maps our principal commitments to the recognised frameworks.

Mapping is indicative and provided to aid assurance; it does not assert certification. Where we hold or pursue formal certification (e.g. ISO/IEC 42001), status is available on request.

16. Review, Versioning and Contact

This policy is owned by the House of Wisdom AI Governance Board and reviewed at least annually. Questions, concerns or requests for assurance documentation may be directed to hello@thehouseofwisdom.com.

17. References

• NIST AI Risk Management Framework (AI RMF 1.0) and Generative AI Profile, NIST-AI-600-1 (2024) — https://www.nist.gov/itl/ai-risk-management-framework
• OECD AI Principles (2019, updated 2024) — https://www.oecd.org/en/topics/ai-principles.html
• UNESCO Recommendation on the Ethics of Artificial Intelligence (2021) — https://www.unesco.org/en/artificial-intelligence/recommendation-ethics
• ISO/IEC 42001:2023 — Artificial Intelligence Management System — https://www.iso.org/standard/42001
• EU AI Act, Article 14 — Human oversight — https://artificialintelligenceact.eu/article/14/