What does Slack actually do with your messages?
Slack's privacy posture in concrete terms — who can read your messages, what's used for AI, what gets retained, and what would be produced under a legal demand.
Slack is the workspace messaging standard at most companies, used by a meaningful fraction of office workers globally for both work and ambient personal-life coordination. The question of what Slack actually does with your messages is concrete, mostly answerable from public docs, and worth understanding precisely if you're using Slack for anything sensitive.
The short version: Slack stores your messages in cleartext on its infrastructure, its engineering team has technical access, its AI features can read your content, and a legal demand for content can be honored. None of this is malicious — it's the standard SaaS posture and the architecture Slack chose to enable its product features. But it's specific, it's verifiable, and it's different from what end-to-end encrypted alternatives like Signal or Koaich do.
Where messages are stored
Slack messages are stored in Slack's databases (running on AWS infrastructure in the United States, plus regional residency options for Business+ and Enterprise customers). Encryption-at-rest uses AWS KMS-managed keys. Encryption-in-transit uses TLS 1.2+ between client and server.
The data itself is stored as readable rows — channel ID, sender ID, timestamp, message text, file references. Files attached to messages are stored as objects in S3-equivalent storage. The reason it's readable is that every product feature you use — search, threading, AI summaries, notifications, integrations — requires Slack's servers to operate on the content.
Who at Slack can read your messages
Slack's standard posture: engineering and support staff have technical access to customer content; access is governed by policy, role-based access controls, and audit logs. Customers on Enterprise Grid get more detailed access-logging and can opt into Slack EKM (Enterprise Key Management) for additional control. Even with EKM, Slack's runtime processing still needs cleartext access.
What this means in practice: if a Slack engineer needs to debug a customer issue, they can read the relevant channel. If a Slack support agent investigates a workspace bug, they can examine the affected messages. The access is logged but exists — it's a property of the architecture, not a temporary state.
What Slack AI does with your messages
Slack AI (the paid add-on) reads workspace messages and files to generate summaries, search results, and recap features. It operates within the customer's workspace; Slack has stated it doesn't train its general models on customer data. The architectural property: Slack AI needs cleartext access to function, which means Slack's servers have cleartext access to function.
Defaults vary by plan; workspace admins can opt out of Slack AI grounding. If you've never deliberately opted out and you're on a tier where Slack AI is included, your messages may be used as AI grounding context. Check the workspace's AI settings at slack.com/admin/settings.
Retention and what gets kept
Slack's default is to retain all messages and files indefinitely. Workspace owners on paid plans can set retention policies that delete messages after a configurable window (7, 30, 90, 365 days, or custom). The policy applies going forward; messages already in storage at policy-change time follow the new rule.
On the free tier, Slack retains the most recent 90 days of message history; older messages aren't accessible to the workspace but Slack's policy on whether they're physically deleted from infrastructure is less transparent.
What's produced under a legal demand
Slack receives subpoenas, search warrants, and other legal process requests routinely (Slack publishes a transparency report). Because Slack holds the keys and stores cleartext, what Slack can produce in response is: messages, files, channel membership history, account metadata, IP logs, and any other data tied to the requested workspace or user.
This is the structural property that motivates the Koaich-style architecture: a workspace tool that holds the keys can produce content; a workspace tool that doesn't can only produce metadata. Slack's privacy policy is a policy-level statement of how Slack will respond; it's not a cryptographic guarantee about what's possible.
If this matters to you
For most workplace conversations, Slack's standard posture is fine — the same way Gmail's standard posture is fine for most email. The question is which conversations are sensitive enough that vendor-side cleartext storage is the wrong shape.
For client conversations under privilege, draft contracts, pre-announcement business intelligence, source identities, or therapy notes, the structural answer is to use a tool that holds ciphertext, not cleartext. How Koaich compares to Slack →