RM-R0017 · ChatGPT · Claude · 1 minute · this page costs 50p read to the end

Nothing is sent in your name until you have seen it

It replies on your behalf, to the wrong person, with the wrong attachment, or in a tone you would not use. Sent mail cannot be recalled.

What you get: Nothing is sent in your name until you have seen the words and the recipients.

Tell the agent.
Paste this line into ChatGPT's custom instructions, or a Claude project's instructions; in Claude Code, CLAUDE.md. It is the rule in the agent's own words; on its own it is an expectation, which the agent follows when it can.
```
Never send an email, message or invitation on my behalf without showing me the exact text and recipients and waiting for a yes; drafts are fine.
```
There is no hook here, and that matters.
ChatGPT and Claude take the line and nothing else, so this behaviour policy is an expectation and the page says so. The stronger controls riskmandate.ai lists are below; in Claude Code, the same line goes in CLAUDE.md.
Try it.
Paste this, as you would, and watch what happens.
```
Tell the team the release is done and thank them. Send it from my mail.
```
With the behaviour policy: You get a draft, the recipients and the subject, and a question: send it? Nothing goes until you say yes to that message.
Without it: The message goes, to the recipients the assistant chose, in the words it chose.
ChatGPT
1. Custom instructions: Never send an email, message or invitation on my behalf without showing me the exact text and recipients and waiting for a yes; drafts are fine. (RM-R0017)
2. Tell the team the release is done and thank them. Send it from my mail.
3. Here is a draft. To: platform-team. Subject: The release is out. "Thank you all: the release went out this morning, on time and without a rollback."
4. Shall I send exactly this, to exactly these people? I'll wait for your yes.
5. Add Priya, then send.
6. To: platform-team, Priya. Same text. Send it now?
7. Yes.
8. Sent.
The send-as-you line in ChatGPT's custom instructions: a draft, the recipients, and a question, and nothing sent until you say yes to that message. An illustration, not a recording.

What this is, and what it is not

Half	Barrier	What holds it
The line	expectation	the agent reads it and follows it when it can

Neither half is a boundary, the only barrier that holds when the agent does not cooperate. The stronger controls for this rule, from riskmandate.ai:

setting An organisation's admin can set a claude.ai connector tool to ask.
Set by the organisation's admin for each connector tool; the setting also reaches Claude Code sessions.
Source: https://code.claude.com/docs/en/permissions, read 2026-10-08, as riskmandate.ai quotes it.

Tell us what happened

Did it stop what it should, and nothing else? What did it get in the way of? Write to agent@riskmandate.ai; it goes on your follow-up, and a person reads it within a working day.

The rule is RM-R0017 on riskmandate.ai, version 1.0.0, CC BY 4.0; copied here on 2026-10-21. The hook and this page are this site's, CC BY 4.0 and Apache-2.0.