An automated reviewer that cannot distinguish "I have no grounding" from "this passes" will emit false approvals. A gate must be able to abstain — grounded-or-silent — and a contract breach must fail the check, not pass it.
Give an automated reviewer two options — approve or reject — and you have built a system that approves under uncertainty. The model that cannot say "I lack the grounding to judge this" says the next-most-available thing, and in most tooling that resolves to letting the change through. Silence becomes consent.
A reviewer is only a gate if it can refuse to answer. Grounded-or-silent: it speaks when it has the evidence to judge, abstains otherwise, and an abstention or a self-contradictory verdict fails the check rather than passing it. Without that, an automated reviewer is a confidence generator, not an enforcement point.