Responsible AI
AI assists. Humans decide.
AI inside PAS solutions does not make compliance, legal, or final financial decisions on its own — these are the principles that govern how it is used.
AI assists, humans remain accountable
Technology supports a decision; it does not replace accountability for it. PAS designs AI-enabled solutions, according to the agreed project scope and the nature of the decision, to maintain a clear human owner for decisions with real effect.
Human oversight
Sensitive decisions pass through human approval before execution. AI can suggest, summarize, and flag — but final approval of consequential decisions remains human, under the institution's approved policy.
Purpose limitation and data minimization
Every AI use is tied to a defined, authorized purpose, and is given only the minimum data required for that purpose — never open-ended access to all of an institution's data.
Traceability
Where AI contributes to a material recommendation or decision, the solution is designed to provide an appropriate level of traceability — what data it drew on, when, and under what rule — according to that project's scope.
Governed use, not open-ended
AI operates within a pre-configured, bounded scope for each solution — not a general open model left to decide freely outside those boundaries.
Validation and context-appropriate explainability
AI outputs are validated before being used in a consequential decision, and are explained to a degree appropriate to that decision's nature — sensitive decisions receive a higher level of explanation than routine suggestions.
What AI does not decide alone
AI inside PAS solutions does not, on its own, make: compliance decisions, legal decisions, Re-KYC customer-restriction decisions, government eligibility decisions, or final financial decisions. Any action or restriction is executed according to the institution's approved policies, configured rules, customer status, and applicable regulatory requirements — not an independent decision by the system.
Security
Security around AI components — access, secrets, logging — follows the same approach described on the Security page.
Changes to this page
Version 1.1 — effective as of 2026-09-15. When a material change is made to this page, the effective date above will be updated.