Slack is reportedly using customer data to power machine learning (ML) features, such as improving the relevance and ranking of search results, sparking criticism over a confusing policy update that led many to believe the company's data was being used to train AI models.
According to the company's policy, anyone who wants to opt out should ask their organization's Slack administrator to email the company to stop using their data.
The revelation came after Duckbill Group executive Corey Quinn asked X, “Sorry, but what the hell are you doing with Slack, your DMs, your messages, your files, etc?”
![(Corey Quinn's tweet and Slack's response):Corey Quinn (@QuinnyPig) tweeted on May 16: "Sorry, Slack. [expletive] What happens to users' DMs, messages, files, etc? I'm pretty sure I'm not reading this correctly." Attached is a screenshot of Slack's opt-out instructions for their machine learning models: Slack (@SlackHQ) responded with the following: "Hello from Slack! To be clear, Slack has platform-level machine learning models for channel and emoji recommendations, search results and more. And yes, customers can opt out of their own data to help train those (non-generative) ML models. Your data is yours. We don't build or train these models to learn, memorize or recall any piece of customer data. Our privacy principles applicable to search, learning and AI can be found here. https://news.google.com/rss/articles/CBMiYGh0dHBzOi8vcmVhZHdyaXRlLmNvbS9zbGFjay1mYWNlcy1iYWNrbGFzaC1vdmVyLXVzZXItZGF0YS11c2VkLWZvci1haS1tYWNoaW5lLWxlYXJuaW5nLXRyYWluaW5nL9IBAA?oc=5." Another response from Slack (@SlackHQ) on May 17: "Slack AI is a generative AI experience built natively into Slack and is a separately purchaseable add-on that uses large language models (LLMs), but does not train those LLMs on customer data. Slack AI hosts its models on its own infrastructure, so data is stored in Slack’s database. https://news.google.com/rss/articles/CBMiYGh0dHBzOi8vcmVhZHdyaXRlLmNvbS9zbGFjay1mYWNlcy1iYWNrbGFzaC1vdmVyLXVzZXItZGF0YS11c2VkLWZvci1haS1tYWNoaW5lLWxlYXJuaW5nLXRyYWluaW5nL9IBAA?oc=5."](https://readwrite.com/wp-content/uploads/2024/05/Screenshot-2024-05-21-125728.png)
Quinn was referring to an excerpt from Slack's privacy principles, which state: “To develop AI/ML models, our systems analyze Customer Data submitted to Slack (e.g. messages, content, files) and other information (including Usage Information) as defined in our Privacy Policy and Customer Agreements.”
Another excerpt reads: “To opt out, please have your organization, workspace owner or primary owner contact our Customer Experience team. [email protected]. “
Slack quickly responded to the post, confirming that it uses customer content to train certain AI tools within the app, but clarified that this data isn't used for premium AI services and that it's completely separate from user information.
The cloud-based team communications platform told TechRadar Pro that the information it uses to power ML is anonymized and it does not access messages.
Meredith Whitaker, president of end-to-end encrypted messaging app Signal, criticized Slack's use of data, saying on X, “Signal would never do anything like this. We don't collect any of your data in the first place, so there's nothing to 'mine' for our 'AI.'”
PSA: Signal would never do this ❤️😇
We don't collect your data in the first place, so there's nothing to “mine” for our “AI” https://t.co/DMyuaTFP33
— Meredith Whittaker (@mer__edith) May 17, 2024
Slack responds to criticism over training models with user data
Responding to the community uproar, the Salesforce.com-owned company posted another blog addressing concerns, saying, “We do not build or train these models to learn, remember or reproduce any kind of customer data.”
Slack also confirmed that user data will not be shared with third-party LLM providers for training purposes.
A Slack engineer attempted to clarify the situation in Threads, explaining that the privacy rules were “originally written for the search/recommendation work we've been doing for years, pre-Slack AI,” and acknowledging that they “needed to be updated.”
Featured Image: Canva
