Responsible AI
Autonomy is granted, not assumed.
This page states how we keep humans in control of systems that act, what we commit to operationally, and the work we will not take. It is written to be held against us.
01Operating principles
Six rules that constrain every system we ship.
These are architectural constraints, not values statements. Each one shows up as something a system does or refuses to do at runtime.
01
Human control is structural
Every system we build has a defined human owner, an explicit action envelope and a path to intervene mid-loop. Control is expressed in the architecture, not in a policy document written after launch.
02
Evidence before autonomy
Scope expands only when the previous scope has produced measured, reconstructable outcomes. No system earns a wider envelope because it feels reliable.
03
Traceable by construction
Context, reasoning, policy version, actor identity and effect are captured at the moment of action, not reconstructed later from logs that were never designed for the question.
04
Model-agnostic by design
No system is architecturally dependent on a single provider. Models are replaceable components behind an evaluation harness, because capability and pricing will keep moving.
05
Reversibility as a design constraint
Where an action can be undone, the undo path ships with it. Where it cannot, irreversibility raises the approval bar rather than being noted as a risk.
06
Honest status
We label what is live, in pilot, in development and conceptual, on this site and in every engagement. Nothing is described as more real than it is.
02What we commit to
Operational commitments.
Data provided for an engagement is used for that engagement. We do not train shared models on client data.
Any system we operate can produce, on request, a full account of a given action within the retention window agreed for that engagement.
Where a decision materially affects an individual, a human review path exists and is documented.
Model, prompt and policy versions are recorded with the actions they produced, so behaviour can be attributed to a specific configuration.
Security posture, access scope and data residency are agreed in writing before a system is granted the ability to act.
03What we refuse
Work we will not take.
We do not build systems that make consequential decisions about people without a documented human review path.
We do not deploy autonomous action into a process we cannot instrument or reverse.
We do not claim customers, certifications, results or scale that we cannot evidence.
We do not take engagements where the objective requires the system to obscure that it is a system.
We do not widen an action envelope because a stakeholder is impatient with the evidence.
The thesis
The next generation of software will not wait to be asked. It will understand, decide, act, and prove what it did.
We’re building that generation from Dubai’s DIFC, for organisations that have to answer for what their systems do.