What can an AI agent do with
Start direct messages?
Lets the agent open a direct message conversation with people in the workspace. Posting the message itself also needs chat:write.
How to spot it
On a Slack connection or settings screen, this is im:write.
With it, an agent can
- Open a direct message with a person
It cannot, on its own
- Read DMs (that needs im:history)
What a bad day looks like
Paired with chat:write, a tricked agent can DM many coworkers with a convincing but false request.
Set it up safer
- Require your approval before it messages someone for the first time
- Ask whether it can be limited to replying rather than starting new DMs
Ask the company selling the agent
- Who will the agent start DMs with, and why?
- Can I limit it to replying only?
Source
Checked against Slack documentation on 2026-09-28. All Slack permissions →
Thinking about a specific agent?
Tell us what it connects to, what it is allowed to do and who approves its work. You get a plain-language label of what it can touch and what to ask the seller. It takes a couple of minutes, and your answers stay in your browser unless you choose to post a test job.
Or have a person go through the AI apps on up to 3 of your business services with you and get a written report, for $249.
This page explains what a permission allows. It does not describe, test or rate any product. Permission details are summarized in our own words from the provider's documentation, linked on the page; providers change their documentation, so check the source before you rely on a detail.