Does WebGPU expose canvas frames to the accessibility tree by default?

Goal

Determine whether rendered WebGPU canvas frames are exposed to assistive technologies by default, or whether all accessibility semantics must be authored in the surrounding DOM.

Constraints and uncertainty

WebGPU renders into an HTMLCanvasElement as a bitmap surface. The specification does not assign an implicit ARIA role, screen-reader label, focus order, or live-region behavior to that surface. Canvas accessibility text is not stable across implementations, device-loss reporting is implementation-dependent, and this behavior requires current verification across implementations.

Assuming a current WebGPU implementation in a modern browser, does an unadorned canvas appear in the accessibility tree? Which wrapper attributes or ARIA roles are necessary for screen readers to discover frame content?