What Beverly will not do

The list, in one place, of things Beverly refuses. Some are not built yet. Some will never be built.

LiveWorking today on every Beverly account.

Every AI product has a boundary. Most of them make you find it. Here it is.

Refusals, and which kind they are
Beverly will not	Why
Send anything you have not approved, unless you granted a standing rule covering exactly that kind of routine message	By design, permanently.
Let a standing rule cover any message to a lead	By design, permanently. A lead has agreed to nothing, so every message to one needs you on the words.
Negotiate, or commit you to anything	By design, permanently. Beverly hands these to you.
Give legal advice, or pass on somebody else's	By design, permanently.
Touch anything involving money movement or wire instructions	By design, permanently, and enforced in code rather than by asking a model nicely.
Send one message to a list of people	Removed from the product deliberately. Not a gap, and not queued.
Text a list you bought or scraped	Refused at the point you record where the list came from.
Answer an inbound phone call	Not built.
Tell a lead anything about what we know about them	By design, permanently. What we know shapes which conversation to have, and then it stays out of the conversation.
Claim a capability it does not have	The list of what Beverly can and cannot do is generated from the code that would run it rather than written by hand, so it cannot drift into optimism.

When Beverly is not sureThe default is to stop and ask, not to try. An assistant that guesses when it is uncertain is only useful until the first time it guesses wrong about something that mattered.

Checked against the product on 29 August 2026. What the labels mean

Language