Slack under fire for sneaky AI training policy

Slack is facing intense backlash from developers and enterprise customers after users realized the platform is using their chat data to train machine learning models by default.

To make matters worse, anyone who wants out cannot simply toggle a switch in their admin settings. They have to compose and send an email to Slack to request an opt-out.

A quiet clause comes to light

The storm began when a post on Hacker News linked directly to Slack's privacy principles page without any extra commentary. The thread quickly went viral as users realized the platform had opted them in automatically.

According to records on the Internet Archive, these terms have been in place since at least September 2023. Yet most workspace administrators had no idea the policy even existed, let alone that it required an email request to escape.

For a company owned by Salesforce that heavily markets customer data control, forcing users through manual email hoops to protect their conversations sparked immediate outrage across social platforms.

Global models versus Slack AI

Slack quickly pushed back against the worst assumptions. A company spokesperson clarified that customer data is only used to train platform-level machine learning models, referred to in the policy as global models, for features like search results, channel recommendations, and emoji suggestions.

Slack insisted that these systems are not built or trained to memorize or reproduce customer data. The company also stated that users who opt out still get access to the benefits of those globally trained models.

Crucially, Slack stated that its newer paid add-on, Slack AI, does not train underlying large language models on customer data. Those language models run directly on Slack's private AWS infrastructure, meaning no data leaves for third-party model providers.

However, none of this nuance was clear in the public policy. In response to criticism on Threads from engineer and author Gergely Orosz, Slack engineer Aaron Maurer admitted that the terms were written before Slack AI launched and need to be updated to clearly explain how the rules apply today.

Why it matters

For teams and workspace admins, this is an immediate operational headache. If your organization handles proprietary code, medical records, or strict client confidentiality, you likely need to draft that opt-out email right away to ensure your internal discussions stay completely untouched.

For the broader software market, this controversy highlights the growing friction between enterprise SaaS platforms and generative AI training. Companies are racing to ship intelligent features, but customers are far more vigilant about how their proprietary information is being harvested.

We think Slack handled this poorly. Hiding data collection under outdated terms and forcing users to opt out via email feels like a relic from an older era of tech. When trust is the primary currency for enterprise collaboration tools, making privacy controls friction-heavy only alienates the very developers and businesses paying the bills.