A structurally different data-exposure problem
Secure multi-party computation lets several parties jointly compute a result from their combined private inputs without any party ever seeing another's raw data — each contributes their own input, the computation runs across all of them cryptographically, and only the final output is revealed, with no party's underlying data exposed to any of the others, or to the process itself, at any point along the way.
This is a fundamentally different shape of problem from the one most access control is built to solve, which generally assumes some party does hold the sensitive data in the clear and has to be prevented from misusing or over-sharing it. Here, the raw inputs are never held in the clear by anyone beyond their original owner in the first place — there's no party in the middle whose access needs restricting, because no such party exists in the computation at all.
What the authorization question becomes instead
Because the inputs are never exposed, the meaningful authorization question shifts entirely to the output: who may receive the computed result, and what could that result, on its own, reveal about any individual input that contributed to it. A result computed this way can still leak information about a specific input under the right conditions, which means output access still needs real, deliberate scrutiny — the technique changes where the risk concentrates, it doesn't eliminate it.