Important Note: Available in Preview for selected customers only. If you are interested in configuring AI Guardrails while in Preview, please contact your account team.
Table of Contents
- Overview
- How it Works
- Checking if AI Guardrails are Enabled
- Configuring AI Guardrails
- Supported Products
Overview
Generative AI Guardrails are essential safeguards designed to ensure the responsible and ethical use of AI technologies within the platform. They act as a regulatory method, including data privacy measures, content moderation, and accuracy verification, to prevent the generation or spread of harmful, inappropriate, or inaccurate information. By implementing these guardrails, we aim to balance innovation with accountability, making sure AI outputs align with ethical standards and user safety.
How it Works
AI Guardrails work by applying specific types of checks to AI interactions:
- Jailbreak Prevention: Keeps the AI operating within its intended safe boundaries and prevents users from bypassing built-in safety, ethical, or legal constraints.
- Toxicity Prevention: Filters out harmful, abusive, or offensive language (e.g., racist, sexist) to maintain a respectful and safe communication environment.
Talkdesk AI Platform currently supports both Jailbreak Prevention and Toxicity Prevention guardrails.
Checking if AI Guardrails is Enabled
You can verify if AI Guardrails is active for your account within AI Trainer. To do so, please follow these steps
- Navigate to the Generative AI section [1].
- Click on Guardrails [2].
- If enabled, you will see a screen confirming "You are protected" [3]. When enabled for an account, all currently supported guardrail types (Jailbreak and Toxicity prevention) are activated together.
- If disabled, a message indicating "Guardrails protection is currently paused" [4] will be shown.
Supported Products
Currently, AI Guardrails supports:
Note: Support for additional products and features will be added in future releases.