The Slack AI Guardrails are a set of protections for all native Slack AI features. To ensure that everyone with access to AI features in Slack can use them safely, the guardrails include several protective mechanisms: 

• Content thresholds to prevent hallucinations
• Prompt engineering with explicit safety instructions
• Context engineering to reduce the risk of prompt injection
• URL filtering to prevent phishing attacks
• Output format validation
• Content safety filters

How the guardrails work

Our guardrails are an additional layer of protection on top of the safety measures built into the large language models (LLMs) all AI features in Slack rely on. They include prompt-side safety measures that are invoked before requests are sent to an LLM to ensure responses are safe and secure.

If a request or response doesn’t pass the guardrails’ safety checks, Slack may modify or block requests and generated content to reduce harm. In most cases, users are guided back to safety via AI-generated responses. For example, if you were to ask Slackbot who your most unproductive coworker is, it’ll tell you it can’t answer and may suggest asking a different question. 

Manage content safety filter settings

Our content safety filters apply to Slack AI features that rely on user-generated inputs to mitigate risks associated with misuse and malicious activity:  

• Slackbot
  Your personal agent for work, built right into Slack.
• Canvas content creation
  Use AI to create a canvas from scratch or add content to an existing canvas.
• Search
  Ask a question and get an answer from AI instead of sifting through search results.
• Workflow creation
  Describe a task or process you want to automate and AI will create a workflow for you.

Owners and admins can adjust settings for content safety filters to comply with internal policies and regional regulatory requirements, or to prevent blocking of legitimate AI use (like asking policy questions or searching for security information). 

Available settings

Here are the available settings:  

• Maximum
  Intended to block targeted employee profiling like performance and protected class information, in addition to the default content safety filters. 
• Default
  Intended to block prompt attacks, violent language, hate speech, sexual content, and other illegal activity.
• None
  No additional content safety filters applied. All other Slack AI Guardrails safety measures remain in place if you select None.

Manage settings

Workspace Owners and Admins (on Pro, Business+, and Enterprise Select) and Org Owners and Admins (on Enterprise Grid and Enterprise+) can manage content safety filter settings.