An obligations register turns regulation from prose into rows. Each row is one thing the firm must, shall or should do, with the clause it comes from, the entity and business line it binds, the date it takes effect, the person who owns it, and the policy and control that satisfy it. A 200-page regulatory notice typically yields a few dozen rows.
The register is what makes compliance auditable. Without it, a supervisor’s question “show me how you meet this requirement” is answered by searching policies for keywords and hoping. With it, the answer is a row, a control and the evidence that control produced. It is also where regulatory change lands: a new rule adds or amends rows, and any row with no control behind it is a gap with a name and a date on it.
Most registers fail on maintenance rather than design. They are built once for a licence application or an audit, then drift as rules change and controls are retired. Language models are well suited to the extraction step that keeps a register current, provided every row they draft is reviewed against the clause before it is accepted.