Slack confirms it is training some of its AI-powered features, but not its generative AI tool, on user content and uploads, with all users opted-in by default
if you don't want your data to be used for training, then take sovereignty of it @slackhq : @QuinnyPig Hello from Slack! To clarify, Slack has platform-level machine-learning models for things like channel and emoji recommendations and search results. And yes, customers can exclude their data from helping train those (non-generative) ML models. Customer data belongs to the Elizabeth Wharton / @lawyerliz : Wow @SlackHQ - what a fine #privacy mess you've created here. You're doing it all wrong. It should be #privacybydesign yet you've f*cked it all up. Corey Quinn / @quinnypig : I'm sorry Slack, you're doing fucking WHAT with user DMs, messages, files, etc? I'm positive I'm not reading this correctly. [image] Rat King / @mikeisaac : slack says it is going to (or already is/has) scan customer data and messages to train their AI models — customers can opt out (rather than opt-in) is this a new development? seeing this link going around and spotted by @QuinnyPig — seems....aggressive https://slack.com/... [image]
Context & Ripple Effects
Slack’s disclosure draws a line between its generative AI tool and platform-level models used for recommendations and search, but places the burden on customers to exclude their data from the latter. That makes training permissions a product-governance issue for workplace administrators, not only an individual privacy setting.
The dispute follows a broader shift in collaboration platforms’ AI terms: Zoom had already limited its stated use of customer content without consent in its AI-training terms update, while LinkedIn later adopted a default training setting with an opt-out path. Slack is an important case because its service concentrates internal conversations and uploaded work material.
First-order effects
- Slack customers must assess whether their workspace data is contributing to non-generative models and use the available exclusion mechanism if that conflicts with their internal data policies.
- Slack can continue improving channel, emoji and search recommendations with eligible customer data, while maintaining its stated boundary against training its generative AI tool on user content and uploads.
Second-order effects
- Enterprise buyers and privacy teams gain another AI-training term to compare across collaboration vendors, raising the value of clear admin controls, model-specific disclosures and contractual assurances.
- The contrast with Zoom’s consent-oriented limitation in its customer-data AI policy puts pressure on vendors to distinguish operational ML features from generative AI rather than treating “AI” as one undifferentiated category.
Third-order effects
- If default inclusion remains common, data-governance choices will increasingly be embedded in SaaS procurement and workspace administration, with opt-out design becoming a competitive trust variable.
- The durable fault line is likely to be purpose-specific permission: customers may accept data use for search or recommendations differently from use in generative models, forcing platforms to make those boundaries legible and controllable.
The trend: Enterprise software is turning customer content into an inference input while renegotiating which model uses require explicit consent versus administrator opt-out.