To make this work, I suppose it will finally be necessary for Windows to disallow all user-space code injection (e.g. in-process hook DLLs), including from assistive technologies. I guess this tightened security could be a per-app opt-in feature, at least initially. UI Automation on Windows 11 may finally be ready to take over the work that in-process injected DLLs (particularly from screen readers) previously did without performance regressions, though as far as I know, this hypothesis hasn't really been tested yet (or if it has, that happened inside the Windows accessibility team at Microsoft after I left). The trick will be to give the third-party screen reader developers a strong incentive to prioritize moving away from third-party code injection, without harming end-users in the process (i.e. not suddenly releasing a browser or OS update that breaks web browsing with screen readers).
What other changes or API additions do you think will be necessary to enable workable app-level RA on Windows?