Slack artificial intelligence, machine learning, and data usage

This guide is a supplemental resource that provides detailed information about the data practices Slack uses in building artificial intelligence (AI) and machine learning (ML) features.

Tip: We recommend reading Our approach to AI and machine learning before reviewing the more detailed information below.

Data overview

As defined in our Privacy Policy, Customer Data is comprised of the following:

• Message data, ex. the content of a message
• File data, ex. files uploaded to Slack
• Object data, ex. a channel or a list
• Transcription data, ex. the transcript from a huddle

Slack does not use Customer Data to train large language models (LLMs) used in generative AI features. As defined in the Slack Privacy Policy, our systems may analyze usage information (like how often a feature is used or interacted with) and workspace information (like the number of users in a workspace or workspace settings) for ML features.

Types of models

We use a variety of models to power Slack’s AI and machine learning features. Understanding these models and how they work is key to understanding how your data is used.

• Generative models
  These models use third-party LLMs to create an output. Customer Data is not used to train these models.
• Predictive models
  These models use machine learning algorithms to power features like channel recommendations and emoji suggestions. Most of our predictive models are global models, meaning they are trained on aggregate data from multiple customers.

How features use models and data

The table below outlines the data and models that power specific Slack features.

Feature	Examples*	Data used	Generative model used?	Predictive model used?
Summarization	Channel summaries, file summaries, huddle notes, recaps	Customer Data: Message, file, and transcription data Other information: Workspace, account, and usage information	Yes, ex. to generate a summary Note: This model never uses Customer Data for training	Yes, ex. to personalize a summary Note: This is a global model, but no Customer Data is used for training
Search	Natural language search, search ranking	Customer Data: Message, file, transcription, and Slack object data Other information: Workspace, account, and usage information	Yes, ex. to generate a search answer Note: This model never uses Customer Data for training	Yes, ex. to rank search results Note: This is a global model, but no Customer Data is used for training
Recommendations	User search, channel search, channel archive, join, and leave suggestions, VIP	Customer Data: N/A Other information: Usage information	No	Yes, ex. to make recommendations Note: This is a global model, but no Customer Data is used for training
AI actions	AI workflow creation, summarize channel workflow step	Customer Data: Message, file, transcription, and Slack object data Other information: Workspace and account information	Yes, ex. to generate workflows from natural language queries and power summary and search workflow steps Note: This model never uses Customer Data for training	No
Autocomplete	@mention recommendations, search navigation suggestions	Customer Data: N/A Other information: Workspace, account, and usage information	No	Yes, ex. to produce recommendations Note: This is a global model, but no Customer Data is used for training
Spam	Spam detection models to prevent abuse of the platform	Customer Data: N/A Other information: Workspace, account, and usage information	No	Yes, ex. to classify spam Note: This is a global model, but no Customer Data is used during training
Translation	Language detection, AI language translation	Customer Data: Message, file, and transcription data Other information: Workspace, account, and usage information	Yes, ex. to generate translations Note: This model never uses Customer Data for training	Yes, ex. to detect language for translations Note: This is not a global model

*The Examples column of this table is included for illustrative purposes and does not comprise an exhaustive list of Slack features.