Slack has updated its AI principles after a controversy emerged over how its customer data was being used to train its machine learning models.
The story broke after an executive at DuckBill Group, Corey Quinn, posted about Slack’s Privacy Principles as they were last week on X. Quinn highlighted that Slack was training its machine learning models on user data and that users have to explicitly opt out of the process.
This can only be achieved by users asking their organisation's IT admin to contact Slack as the company representative to ask it to stop. The admin must email [email protected] with the organisation's Workspace/Org URL and the subject line "Slack Global model opt-out request".
Quinn said, "I'm sorry Slack, you're doing f***ing WHAT with user DMs, messages, files, etc? I'm positive I'm not reading this correctly."
Slack has said it employs machine learning to underpin in-app features such as channel recommendations, search results, autocomplete, and emoji suggestions. However, the suggestion that it utilises users' Slack messages, data, and files to enhance these features led to frustration among users, especially those who were unaware that they were automatically opted into this policy.
On Friday, Slack posted a blog seeking to clarify the situation and outline what specific data is used for training models and which isn't, as well as how the customer data used is being treated.
Slack's blog wrote:
We do not build or train these models in such a way that they could learn, memorize, or be able to reproduce any customer data of any kind. While customers can opt-out, these models make the product experience better for users without the risk of their data ever being shared. Slack’s traditional ML models use de-identified, aggregate data and do not access message content in DMs, private channels, or public channels."
Additionally, Slack stressed that customer data is not used to develop large language models (LLMs) or other generative models, while its add-on generative AI product, Slack AI, leverages third-party LLMs.
Slack emphasised that the machine learning models "make the product experience better" by honing channel and emoji recommendations and search results.
When Engadget approached Slack for comment, a Slack spokesperson said: "We do not build or train these models in such a way that they could learn, memorise, or be able to reproduce customer data.”
Slack also responded to Quinn's post: “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.”
