> use of AI has indeed lowered the story point cost

It should not have. At least not significantly. Points should represent complexity, risk and overall effort (review burden, testing burden, dependencies, etc.), and so AI should increase velocity before it decreases story point estimates.

Over time, if a team's baseline delivery model genuinely changes, then reference stories can be recalibrated, but casually saying "AI lowered the point cost" is usually a smell that points are being treated as time estimates.

This is the same reason points should not = days even without AI. Velocity is what tells you if a team is getting better through training, tooling, hiring/firing, or process improvements. Re-pointing the same class of work downward hides the gain.