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
https://slack.com/... What a time to be alive in IT. 🤦♂️ JP Mens / @jpmens@mastodon.social : “Our guiding principle as we build this product is that the privacy and security of Customer Data is sacrosanct [..] our systems analyse Customer Data (e.g. messages, content and files) submitted to Slack [..] We have technical controls in place to prevent access [..] you can opt out.” — https://slack.com/... … X: @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 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] Ian Kar / @iankar_ : Default opt-in is shady and duplicitous nowadays 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. Meredith Whittaker / @mer__edith : PSA: Signal will never do this ❤️😇 We don't collect your data in the first place, so we don't have anything to “mine” for “AI” https://techcrunch.com/... Simon Willison / @simonw : Slack learning the hard way right now that when it comes to the issue of training AI on customer data people will always assume the worst, and if you try to convince them otherwise they likely won't believe you Rakesh Agrawal / @rakeshlobster : If you're b2b and customers pay you, opt-out is bad form. @tophinity : Wow, @SlackHQ with scumbaggery level turned up all the way. What a piece of shit company @scottmonaghan : God damnit. Let's all go back to paper. “Computers” were a bad idea. Gergely Orosz / @gergelyorosz : Source for the post is Slack https://slack.engineering/... Interesting how Slack claims they do not use customer messages to train LLMs. Yet their own legally binding policies do not rule out that they can do this - in fact, those terms read like they do this! https://twitter.com/... Jamon / @jamonholmgren : My opt-out request to Slack training on our 10+ years of data included some pointed feedback. [image] Michael London / @mlondon83 : This is why I tell my staff to keep any customer sensitive topics out of chat Mark Kember / @markyk : How does this work with companies collaborating on documents under NDA? Every Slack would have to opt out or risk data making its way into training data set. Any company with client confidentiality - law firm, PR firm, accountancy, and more - will have to opt out as standard? David Krueger / @davidskrueger : So... ...you want my data to train your global model... ...but you're training it such that it couldn't possibly learn anything from my data... ...so why do you want my data again?? smells like BS privacy washing at best we have probabilistic guarantees tell us your epsilon! [image] Christopher Stanley / @cstanley : This is a complete breach of privacy! Default on and the only way to opt-out is via an email? @SlackHQ care to comment? @0xjonah1 : Incredible how hard they're making opt-out. “Enclose privacy request in sealed wax envelope, hand off to secret carrier pigeon who appears on block A on Mondays, Wednesdays, Fridays (public holidays off) between the hours of 10am-1pm. Chance of successful delivery 30%. Brandon Goldman / @brandongoldman : This seems sneaky. I don't think companies want all their *private data* used to train Slack's global AI models. Private data is supposed to be private. @Benioff shouldn't this kind of thing be opt-in, not opt-out? Christina Warren / @film_girl : Honestly, I really want the 3800 post Slack thread I started 6 months ago to be in a global training set. It's what the world deserves. The chaos of the danger thread. Mohit Tiwari / @mt_lattices : Orgs have to certify what their customers' data is being used for... Not just through “Terms and Conditions” but by engineering transparency into how they handle data. @curttalkstv : Requiring people to opt-out instead of allowing them to opt-in should be illegal any time potentially private information is involved. @SlackHQ @jermops : “Email us EXACTLY LIKE THIS to opt-out” Congrats to @SlackHQ on achieving Comcast levels of customer service CJ Black / @charleseblack : I've been ready to give up on @SlackHQ for a bit. Are there any decent alternatives? We've tried Teams (bad) and Matrix. Makena Kelly / @kellymakena : it would be kinda sick if congress was able to pass any of those privacy bills that required users to opt-in to this kind of data collection but they haven't :'-) Pepijn Vissers / @purpletheism : If even remotely used for propriatary or slightly confidential stuff, please host your own. Mattermost, for instance. @privacymatters : If your org uses Slack then contact them to opt out of the use of ‘customer data’ to develop their AI/ML models ... why on earth Slack thought this was ok based on an opt out is beyond me .. wither trust https://slack.com/... [image] Jake Williams / @malwarejake : Given that @SlackHQ has built a platform full of integrations, there's only one explanation for requiring a manual email for opt-out: they're intentionally making it difficult. I hate to say it, but this is making Teams look viable. Steven Tey / @steventey : This is wild - but what's even crazier is Slack requires you to send an email to opt-out of selling your data. To make this easier, I created a quick @dubdotco link to craft this email in 1-click → https://d.to/... [image] Puru Dahal / @itspuru : We need an open-source alternative, this time better / cheaper, @Mattermost is not cutting it . Better yet, Discord + Slack alternative (most of the folks/projects building community are on Discord). @GlueAI looks like onto something, wish it was open-source. People flocked to Darian Shirazi / @darian314 : sounds like switching to discord is the move Nikhil / @nik_kothari22 : Not gonna lie, kinda saw this coming. All the more reason for us to build Raven. With open source + hosting it on your own servers/@FrappeCloud you have complete control over how your data is used (or shared - if at all). @kepano : “file over app” is just as relevant for businesses as it is for individuals — if you don't want your data to be used for training, then take sovereignty of it 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] LinkedIn: Saurabh Shintre : Policies like these do no good to enterprise AI adoption. Default opt-ins and then making it hard to opt-out clearly erodes trust in data security … Forums: Beehaw : Slack users horrified to discover messages used for AI training See also Mediagazer
Context & Ripple Effects
Slack’s disclosure distinguishes platform-level machine-learning features from its generative AI tool, but makes customer messages, content and uploads part of the training picture unless customers opt out. It extends a workplace-software pattern seen when Zoom revised its AI-training terms around customer data, while preserving a narrower claim about audio, video and chat content.
The distinction matters because workplace chat already concentrates informal, potentially sensitive communications; earlier coverage connected the growth of these tools with more workplace-harassment disputes tied to messaging. Data-use defaults therefore become a governance issue for organizations, not merely a feature setting.
First-order effects
- Slack customers are opted in by default for training of specified non-generative AI features using their content and uploads; they must use Slack’s available opt-out path to exclude that data.
- Slack can continue improving functions such as recommendations and search with customer-data signals while maintaining that its generative AI tool is not trained on those materials.
Second-order effects
- IT, legal and privacy teams will need to assess Slack’s data-use settings alongside retention, access and workplace-communications policies, especially where uploads and messages contain sensitive business context.
- The disclosure raises the competitive bar for collaboration vendors: clearer feature-by-feature training boundaries and workable customer controls become part of enterprise AI procurement.
Third-order effects
- Enterprise AI is moving toward feature-specific data-use disclosures rather than a simple binary of “AI uses data” or “AI does not,” making defaults and exclusions central product-governance choices.
- If customers increasingly demand affirmative consent, providers may have to shift from platform-wide opt-outs toward more granular controls and clearer separation between operational ML and generative-model training.
The trend: Collaboration software is turning customer content into an input for embedded AI, with consent design and data-boundary transparency becoming competitive enterprise requirements.