After announcing Slack AI in February, Slack faced a backlash, defending its vague policy of siphoning customer data (messages, content, files, etc.) by default to train Slack's global AI model. It seems like it is.
Slack explained in a blog post that the Salesforce-owned chat service does not train its large-scale language model (LLM) on customer data, according to Slack engineer Aaron Maurer. However, Slack's policies may need to be updated “to more carefully explain how these privacy principles relate to Slack AI,” Maurer wrote in Threads. Part of the reason is that this policy was “originally written about the search/recommendation work we've been doing for years before Slack came along.” ”
Maurer was responding to a Threads post by engineer and writer Gergely Orosz, who called on companies to opt out of data sharing until policies are clarified, using actual policy language rather than a blog.
“ML engineers at Slack say they don't use messages to train LLM models,” Orosz wrote. “My response is that the current terms allow them to do so. When it's in a policy, I believe it's a policy. A blog post is not a privacy policy. If you're a serious company. Everyone knows this.”
User tensions become clearer when comparing Slack's privacy principles to the way the company promotes Slack AI.
Slack's Privacy Principles state that “Machine learning (ML) and artificial intelligence (AI) are useful tools that we use in limited ways to enhance our product's mission. To develop AI/ML models. , our system analyzes customer data (messages, content, etc.)” specifically. , files), and other information (including Usage Information) as defined in our Privacy Policy and Customer Agreement. ”
On the other hand, the Slack AI page says, “Work with peace of mind. Your data is your data. We won't use it to train Slack AI.”
Because of this discrepancy, users asked Slack to update its privacy principles to clarify how their data is used in Slack AI and future AI updates. A Salesforce spokesperson said the company agreed that the update was necessary.
“Yesterday, some Slack community members asked us to be more clear about our privacy principles,” a Salesforce spokesperson told Ars. “Today, we are updating these principles to better explain the relationship between customer data and generative AI in Slack.”
A spokesperson told Ars that the policy update will prevent Slack from “using customer data to develop LLMs and other generative models,” or “using customer data to train third-party LLMs.” He said it will be clear that “we will not be building or training these models.” It may be possible to learn, remember, or reproduce customer data. ” The update also clarified that “Slack AI uses an off-the-shelf LLM whose models do not hold customer data,” adding that “customer data never leaves Slack's trust boundary, and the provider of the LLM We guarantee that your data will be inaccessible. Customer data. “
However, these changes don't seem to address key concerns for users who have never explicitly consented to sharing chats or other Slack content used for AI training.
Users opt out of chat sharing with Slack
This controversial policy is not new. Wired warned him about this in April, and TechCrunch reported that the policy has been in place since at least September 2023.
But a widespread backlash began to mount on Hacker News last night, with Slack users criticizing the chat service for not notifying users about the policy change and instead appearing to silently opt-in by default. For critics, it felt like there was no benefit to opting into anything other than Slack.
From there, the backlash spread across social media, with SlackHQ rushing to clarify Slack's terms with an explanation that doesn't seem to address all the criticism.
“Sorry, Slack, what are you doing with your DMs, messages, files, etc.?” Corey Quinn, chief cloud economist at a cost management company called Duckbill Group, told I'm sure I didn't read it correctly.”
SlackHQ responded to Quinn after the economist declared “I hate this” and admitted he had opted out of data sharing in paid workspaces.
“To be clear, Slack has platform-level machine learning models for things like channel and emoji recommendations, search results, and more,” SlackHQ posted. “And yes, customers can exclude their data from training those (non-generated) ML models. Customer data belongs to the customer.”
Later in the thread, SlackHQ said, “Slack AI — a generative AI experience built natively in Slack —”[and] This is a separately purchased add-on that uses large language models (LLMs), but does not train those LLMs on customer data.”
