Since the launch of Slack AI in February, Slack appears to be digging deeper, defending its vague policy that by default siphons customer data — including messages, content, and files — to train Slack’s global AI models.
According to Slack engineer Aaron Maurer, Slack explained in a blog post that the Salesforce-owned chat service does not train its large-scale linguistic models (LLM) on customer data. But Slack’s policy may need updating “to more carefully explain how these privacy principles play with Slack AI,” Maurer wrote in Threads, in part because the policy “was originally written for the search/recommendation work that we performed years before Slack AI.”
Maurer was responding to a post on Threads by engineer and writer Gergely Oros, who urged companies to hold off on sharing data until the policy is clarified, not via a blog, but in the language of actual policy.
“One ML engineer on Slack says they don’t use messages to train LLM models,” Oros wrote. “My answer is that the current conditions allow them to do so. I’ll believe it’s politics when it’s in politics. A blog post is not a privacy policy: any serious company knows that.”
The tension for users becomes clearer if you compare Slack’s privacy principles to how the company advertises Slack AI.
Slack’s privacy principles specifically state that “Machine Learning (ML) and Artificial Intelligence (AI) are useful tools that we use in limited ways to advance our product mission. To develop AI/ML models, our systems analyze Customer Data (eg messages, content and files) sent to Slack, as well as other information (including usage information) as defined in our privacy policy and in your customer agreement.”
Meanwhile, the Slack AI page says, “Work without worry. Your data is your data. We don’t use them to train Slack AI.”
Because of this discrepancy, users have called for Slack to update its privacy principles to make clear how data is used for Slack AI or any future AI updates. According to a Salesforce spokesperson, the company has agreed that an update is needed.
“Yesterday, some members of the Slack community requested more clarity regarding our privacy principles,” a Salesforce spokesperson told Ars. “Today, we’re updating these principles to better explain the relationship between customer data and generative AI in Slack.”
The spokesperson told Ars that policy updates will clarify that Slack does not “develop LLM or other generative models using customer data,” “use customer data to train third-party LLMs,” or “build or train these models in such a way , to be able to learn, remember, or be able to reproduce customer data.” The update will also clarify that “Slack AI uses out-of-the-box LLMs where models do not retain customer data,” ensuring that “customer data never leaves the boundary of Slack’s trust and LLM providers never have access to customer data.”
However, these changes do not appear to address a major concern of users who have never given explicit consent to share chats and other Slack content for use in AI training.
Users who opt out of sharing chats with Slack
This controversial policy is not new. Wired warned about it in April, and TechCrunch reported that the policy is in effect at least through September 2023.
But widespread backlash began to swell last night on Hacker News, where Slack users called out the chat service for apparently failing to notify users of the policy change, instead quietly enabling them by default. To critics, it seemed like there was no point in opting in for anyone but Slack.
From there, the backlash spread to social media, where SlackHQ was quick to clarify Slack’s terms with explanations that didn’t seem to answer all of the criticism.
“Sorry Slack, WHAT the hell are you doing with user DMs, messages, files, etc?” Corey Quinn, chief cloud economist for an expense management company called Duckbill Group, posted on X. “I’m sure , that I’m not reading this correctly.’
SlackHQ responded to Quinn after the economist said, “I hate this so much” and confirmed that he had opted out of data sharing in his paid workspace.
“To clarify, Slack has platform-level machine learning models for things like channel and emoji recommendations and search results,” SlackHQ posted. “And yes, customers can opt out of their data from helping train these (non-generative) ML models. Customer data belongs to Slack AI—which is our generative AI experience built into Slack—[and] is a separately purchased add-on that uses large language models (LLMs) but does not train these LLMs on client data.”
