Responsible innovation in practice

The review, and who decides it

A Responsible Innovation Review is a written assessment prepared by the team making the proposal, examined by the Ethics office in Singapore and decided by the Head of Technology and Ethics. Proposals above US$10 million, and proposals of a kind the group has not reviewed before, go to the full Ethics & Technology Council instead.

The review comes before the model rules, the artificial intelligence rules and the privacy rules, because it governs the decision to begin. A team that has already built a system has already made choices about data and about inference that a review can then only unpick. Reviewing at the point of commitment is cheaper for the group and safer for the people the work touches.

A review ends in one of four ways: cleared, cleared with conditions, deferred for more evidence, or declined. Conditions bind the proposing team or enter the investment documents as covenants, and their fulfilment is verified before launch or before the second drawdown. A deferral is not a soft refusal; three of the deferred proposals since 2024 returned with evidence and were cleared.

Seven questions asked of everything new

The review tests the same seven things whether the proposal is a valuation method, a sector dataset, a client reporting tool or a position in a business developing a veterinary vaccine. Asking the same questions of each keeps the answers comparable across divisions and makes it obvious when one proposal is being held to a lower standard than another.

The questions are written into the review template and answered in order, with evidence attached. An answer of not applicable is permitted only with a reason, and the Ethics office rejects the reason more often than it accepts it. Roughly a third of first submissions are returned for the evidence they omit rather than for the substance of what they propose.

• Who will answer by name for this and for what it does.
• Whether the automation is proportionate to the cost of its errors.
• Where the data comes from and on what lawful basis it is held.
• Whether it infers anything about a person beyond what a mandate requires.
• Whether it affects animals and, if so, whether it meets the Group Welfare Standard.
• Whether a client can see where it was used and require a human decision instead.

What the group builds for itself

New work begins in a development environment separated from production, in which client data is not used unless there is a lawful basis for that specific use and the Information Governance Committee has approved it. Most development runs on masked or synthetic records, and the separation is tested rather than assumed.

No product that makes or recommends a decision about a client is released without the human review point the Charter requires. No product infers a characteristic of an individual that the mandate did not ask for. No group system develops, licenses or uses facial recognition. Every product carries a named owner at head-of-function level and a written description kept current.

What the group funds

Innovation-led positions in the origination sectors involve living systems: medicines and vaccines, biologics grown in cell lines or in animals, breeding and genetic programmes, aquaculture technologies, agricultural inputs. The review for such a position is conducted with the Welfare Assessment Unit and, where research is involved, with an external ethics reviewer engaged under the third-party rules.

The group funds development only where the sponsor holds the approvals the responsible authority in that jurisdiction requires, where any use of animals meets the Group Welfare Standard and where the data supporting the product is held in a form the group can audit. Investment documents require every adverse finding in development to be disclosed to the group within ten working days.

Deployment-sector positions in data, technology, information infrastructure and financial systems face the same seven tests with weight on provenance, on inference about individuals and on the right to a human decision. A technology business selling a product the group would not build for itself is not funded, whatever its growth.

Research the group will not fund

A little research carries a hazard far larger than its commercial value. The group keeps a register of research of concern, approved by the Board on the advice of the Council, and funds nothing on it. The register is reviewed each year with the independent directors and external scientific advisers, and additions are made on evidence rather than on sentiment.

Proposals adjacent to the register, such as a vaccine platform that could be repurposed or a genetic technique with applications beyond the one proposed, receive an enhanced review by the full Council with an external reviewer. They are cleared only with conditions that restrict the use of group capital to the stated purpose and give the group a right to exit if that purpose changes.

• Research whose purpose is to make a pathogen more transmissible or more lethal.
• Heritable genetic alteration of animals for purposes other than health, welfare or disease resistance.
• Systems for the surveillance or profiling of individuals without their mandate.
• Autonomous decision systems in which no named person can be held to account.
• Products whose commercial case depends on inferring a protected characteristic.