AI Principles

Our approach to AI and machine learning

At Slack, artificial intelligence (AI) and machine learning (ML) are core to our product experience. Our commitment to data privacy and security is fundamental to everything we do, ensuring robust controls are in place to protect your information. Slack’s native AI features have been built to uphold these commitments. Your Customer Data (like messages and files) is never used to train any LLMs.

The AI and ML features we build are designed to make working in Slack simpler, smarter, and more efficient. While AI is newer to Slack, ML has long been a key part of the Slack experience. This guide digs into how these technologies are built into Slack, and the clear distinction between them:

• AI specifically refers to generative AI features powered by large language models (LLMs), like channel summaries.
• ML encompasses features using predictive models, like emoji recommendations and display name suggestions when mentioning other users.

AI in Slack

AI features are included in Slack paid plans. Admins can enable or disable AI features, giving customers full control over how and whether AI is used in their workspaces.

Our AI architecture was designed with privacy in mind.

All of our AI features are protected by Slack AI Guardrails, a set of built-in foundational protections. These safeguards include content thresholds that reduce hallucinations, explicit safety instructions that limit prompt engineering, context engineering to reduce prompt injection risks, URL filtering to prevent phishing attacks, and output format validation, as well as provider mitigations.

The Slack AI Guardrails also automatically apply Content Safety Filters as another protective layer for AI features that rely on user-generated inputs. These filters analyze queries in real time to identify and mitigate harmful content, prompt injection attempts, and security risks before they reach AI systems like Slackbot and Search Answers.

Our approach to AI is grounded in these three principles:

• Your data is never used to train large language models.
  Your Customer Data in Slack (like messages and files) are not used to train LLMs.
• You control what AI can access:
  Workspace and organization admins can turn AI features on or off at any time, giving customers full control over how AI is used. Slack AI only works with content you already have permission to view. For example, AI search answers will only include results you could also find in a standard search.
• Your data stays within Slack’s trusted infrastructure.
  The LLMs we use are deployed inside Slack’s cloud environment, so model providers do not have access to your data.

Note: Admins can decide whether members of their workspace or Enterprise organization have access to our AI features. Visit Manage access to AI features in Slack for more details.

Machine learning in Slack

We also use predictive machine learning models to help make the Slack experience even better. When you see an emoji you and your teammates have used recently in the emoji picker, or an autocomplete suggestion to help find the right person at your company with a common name, our ML models are responsible for the relevancy and accuracy of those suggestions.

• Machine learning models allow us to deliver high-quality, personalized experiences.
• We build and train models so that they cannot reproduce Customer Data, and the outputs of the models cannot be linked to data from a specific customer.

Tip: Visit our Privacy Principles for an overview of the principles we follow to inform product development at Slack and Customer controls.

More information on models and data

For more details on the generative and predictive models we use and how they power our AI and ML features, visit Slack Artificial Intelligence, Machine Learning, and Data Usage. This resource provides detailed information for anyone looking for a more thorough understanding of our AI and ML data practices.