Evidence informs.
Separate what was observed from what is inferred. Show the source, uncertainty and relevance to this specific decision.
Governed
intelligence.
We design and create governed intelligence systems. That means making the evidence, conditions and permission behind an action explicit.
Separate what was observed from what is inferred. Show the source, uncertainty and relevance to this specific decision.
Establish the requirements and limits before an action. A confident recommendation cannot cancel an applicable prohibition.
Check who may approve or execute, for which scope and period. Knowledge, confidence and permission are different things.
Keep permission separate from what actually happened. Record attempts, outcomes and unresolved results with an appropriate evidence trail.
Gates are the checks within a system. They establish whether its requirements and authority permit an action.
State the rule and the available remediation or review path.
Identify the missing, stale or conflicting information and who can resolve it.
Present the case to someone with the appropriate authority.
Act only within its conditions, limits and expiry. Re-check before execution.
RAFYL is the standalone AI project being developed towards a later governing role. It does not acquire unlimited authority through that role.
Policies, grants and execution conditions remain explicit. Changes to those boundaries require their own authorised review.
About RAFYLThe interface should explain the next step, expose important uncertainty and keep human choices understandable. The amount of friction should fit the consequence of the action.
A principle describes intended behaviour. A working control needs implementation and testing. Product pages distinguish prototypes, concepts and future integrations.
These principles summarise our development approach. Specific controls and permissions depend on the product and its operating context.