AI Referral Personal-Attestation Consent Gate Automation: The Assistant That Cannot Vouch for You

A business network that positions itself around curation and quality, rather than open sign-up, is making an implicit promise to every member already inside it: the people being let in were personally vouched for by someone who actually knows them. The moment that vouching step can be satisfied by something other than a real person genuinely attesting to it — a form pre-filled with plausible defaults, an assistant answering on autopilot — the promise the entire network is built on quietly stops being true, even if every other part of the admission process still runs correctly.

The tool's own required argument for these four declarations is a single boolean, and the migration's own documentation for that argument is explicit about what it takes to set it: never mark this confirmed unless the account owner has confirmed, using their own words, inside the actual conversation, that they personally know the person, that no payment is involved, that it is not a mass submission, and that they accept the network's policy. That instruction lives directly in the tool's catalog entry — the same documentation an AI agent reads to decide how to use the tool correctly — rather than being enforced only by a separate policy document nobody consults in the moment.

Structurally, the tool has no way to infer or default that confirmation on its own: the argument has no default value that resolves to true, so an agent that skips asking, or that tries to proceed on an assumption, produces a call that the tool refuses outright, with a response naming precisely what confirmation is still missing. This puts the actual point of human judgment exactly where a curated network's trust model needs it — the moment a real person is asked, and answers, whether they personally vouch for someone else, rather than folding that judgment into whatever an assistant infers is probably fine.

A second, related safeguard covers what happens after the confirmation gate is cleared: the referral submission itself is wrapped so that if the network's own underlying rules reject it for an unrelated reason, the tool surfaces that specific rejection message rather than reporting a generic failure or, worse, silently reporting success. An agent handing a rejection back to the account owner can relay the actual reason, rather than leaving them to guess why a referral that should have gone through didn't.

An account owner asks their assistant to "refer someone I met at a conference last week" without further detail. Before this gate, an assistant eager to be helpful might reasonably infer the four declarations are all true and proceed. With the gate in place, the tool has no way to accept that inference — the assistant has to ask the owner directly whether they personally know this person, whether any payment is involved, and whether they accept the policy, and only submit once the owner has actually answered.

A referral into Centriu Vértice's curated network can now be submitted through an AI assistant without weakening the specific human commitment the network's own admission policy depends on — the assistant can prepare and send the referral, but the moment of genuine personal vouching still has to come from the actual person making it.

This gate governs the referral-submission step specifically — it does not decide whether a referred candidate is actually admitted into the network, which runs through the network's own separate, human-reviewed admission process and is unaffected by this tool.

A confirmation requirement documented only as a policy — "agents should confirm before referring" — depends entirely on every agent, every time, choosing to follow it correctly. Centriu's approach removes that dependency: the tool itself has no path to a true confirmation value except one that was actually, explicitly obtained from the account owner in the conversation, which is what makes the safeguard hold even when nobody is separately checking that it was followed.