Ethical Technology
Technology witha named personbehind it.
The Board adopted this Charter in 2023, before the first learning system in the group was put to work. IGUAKO Capital had decided to use models in research, screening, valuation and operations. It decided at the same time that the limits on that use would be written down first, approved by the Board and applied in every division and every jurisdiction without local variation. This document is that writing.
- Charter adopted
- 2023
- Principles
- 7
- Automated client decisions
- 0
- Oversight
- Ethics & Technology Council
Why
The Charter states seven principles. Each principle carries an owner, a control and a test, so that keeping it can be checked rather than asserted. The Ethics & Technology Council administers the Charter from the Singapore hub, approves every use of a learning system before it goes live, and reports to the Board once a year on whether the principles held. The operating text sits in five group policies.
The Charter binds all 28 jurisdictions of the group, all 312 of its people and every vendor that handles group data or supplies a system that learns from it. Only the Board may amend it. No committee, executive or local board may grant an exception to it. Where the law of a jurisdiction requires more than the Charter, the local rule applies. Where it requires less, the Charter applies.
The Ethical Technology Charter
Ethical Technology Charter.
Adopted by the Board in 2023. Seven principles, each with the practice that gives it effect.
Human accountability
A named person answers for every automated decision the group makes, and the name is on the record before the system runs rather than found afterwards. Each of the 71 models in the group inventory has one owner, an individual and not a team. Each of the 26 approved uses of a learning system has an accountable executive drawn from the Executive Committee. Accountability is never transferred to a vendor, to a committee or to the system itself. Where an owner leaves or changes role, a successor is named before the change takes effect or the model is suspended.
- One named individual owner for each of the 71 models in the group inventory.
- An accountable executive from the Executive Committee for each of the 26 approved uses of a learning system.
- A named human author on every decision that reaches a client, a counterparty or an employee.
Proportionate automation
Automation is permitted only where the cost of an error is understood, written down and accepted by a person with the authority to accept it. The model tier records that judgement. Tier 1 covers outputs that determine an order, a reported valuation, a capital figure or a screening outcome, and 12 models sit there. Tier 2 informs a human decision without determining it and holds 26. Tier 3 produces internal information with no external effect and holds 33. Where the cost of an error cannot be stated, the task is not automated at all.
- Three model tiers set by the Chief Risk Officer according to the cost of error, with the Tier 1 list approved by the Board each year.
- A prohibition on any system acting on a client, an employee, a counterparty, a position or a filing without a human decision in between.
- A fifteen-minute stop authority and a tested fallback for every Tier 1 model.
Provenance of data
Every dataset carries a documented origin and a lawful basis before it is used for anything. The processing register held 84 activities at 30 June 2026, and each entry names the categories of data, the origin, the basis in each jurisdiction where the activity runs, the systems that hold it, the retention period and the owner. The group does not buy personal data in bulk, does not take it from public networks and does not accept a dataset whose supplier cannot say where it came from. Data that cannot be traced is deleted.
- An origin statement on every dataset, tested at admission and again at each annual review.
- A privacy assessment before any new processing activity, signed by the activity owner and the Group Data Protection Officer.
- A contractual requirement that every vendor state the origin of any data it supplies.
Restraint in inference
The group draws a conclusion about a person only where a mandate, a legal duty or a contract requires it. Source of wealth, sanctions exposure and political exposure are assessed because client due diligence requires them. Health, beliefs, political opinion, ethnic origin, family circumstances and personal habits are not assessed at all, by a person or by a system. No system in the group scores propensity to any behaviour, builds a behavioural profile or identifies a person from an image, a voice or any other biometric feature.
- Prohibited inferences listed in each register entry and tested before approval with held-out cases.
- Same-day withdrawal of any system found in use to produce a prohibited inference.
- No facial recognition, voice identification or biometric inference in any office or system.
Welfare in biological capital
Welfare standards travel with the capital into every life-science position. The Group Welfare Standard sets six criteria and sits above the legal minimum in every jurisdiction where the group holds a position, so that a business in Brisbane, one in Sao Paulo and one in Johannesburg meet the same floor. No position in scope reaches the Investment Committee without an assessment against the Standard by the Welfare Assessment Unit. Every investment document in scope carries a welfare covenant with the standing of a financial covenant.
- A welfare assessment before commitment on each of the 19 positions in scope, with a site visit above US$5 million.
- A welfare covenant, annual reporting and notification of any incident within five working days in every investment document in scope.
- A Board-approved list of excluded practices, maintained by the Ethics & Technology Council.
Transparency to clients
A client can see which decisions were model-assisted. Every report, valuation, recommendation or communication to which a Class 2 or Class 3 system contributed carries the assistance statement, a fixed form approved by the Council that names the class of system used, what it contributed and the person accountable for the content. The statement does not vary by division, by mandate size or by jurisdiction. A client may ask for the register extract covering any approved use that touched their mandate, and for the welfare summary of any position they hold.
- An assistance statement in a single fixed form on every client output to which a learning system contributed.
- A register extract available on request for any approved use that touched a client mandate.
- Annual publication of the reviews conducted, the approvals given, the refusals and the incidents.
The right to a human decision
Any client may require that a decision be reviewed by a named person who did not take it, whether or not a system assisted it, and may instruct that future decisions on the mandate be taken without system assistance. The review is completed within ten business days and the outcome is given in writing with reasons. Eight such requests were made in the twelve months to 30 June 2026 and two of the original decisions were changed. Employees hold the same right over decisions taken about them.
- A human review by a person who did not take the original decision, completed within ten business days.
- A standing client instruction, recorded in the register, to exclude system assistance from future decisions on a mandate.
- The same right for employees, with no learning system used in recruitment, assessment or remuneration.
Positions
Twelve statements of practice.
Governing artificial intelligence
The group uses learning systems in 26 approved ways and in no others. Each use is entered in a register before it runs, sorted into one of four classe…
ReadThe data the group will hold
Every dataset in the group has a documented origin, a lawful basis, a purpose and a deletion date before anyone works with it. The processing register…
Read
Accountability for automated work
Every automated estimate the group makes has a name attached to it. The 71 models in the inventory have 71 named owners, each an individual. Every app…
Read
How models are overseen
The group ran 71 models at 30 June 2026, tiered by the cost of being wrong rather than by their complexity. Twelve were Tier 1, 26 were Tier 2 and 33 …
Read
Responsible innovation in practice
Novelty is where the advantage sits and where the harm is least understood. Every new product, model, dataset or innovation-led position passes a writ…
ReadPrivacy built into the systems
Privacy in the group is a property of the systems rather than a promise made about them. Fields carry a sensitivity class and a purpose tag. Access is…
ReadWelfare travels with the capital
Nineteen positions sit inside the Group Welfare Standard. Each was assessed against six criteria before commitment, each carries a welfare covenant en…
ReadFinancing medicines on conditions
Pharmaceutical capital carries conditions on manufacture, evidence, supply and price conduct. The group finances capacity, registration and launch wor…
ReadWhere the records sit
The group decides where its records are held, who can reach them and what happens when a supplier relationship ends. Production runs in two environmen…
ReadThe person who signs
Oversight means a signature that could have gone the other way. The group fixes the point at which a person must sign, the conditions that make the si…
ReadWhat suppliers agree to
A commitment the group makes cannot stop at its own boundary. Thirty-eight suppliers handled group or client data at 30 June 2026, and each signed the…
ReadWhat clients are told
A client can see where a system touched their work, ask for the record behind it and require a person to decide instead. The assistance statement appe…
Read