Guardrail Configuration

Philip Cobbold Updated by Philip Cobbold

To configure AI Admin, you must have Customer Administrator permissions, and a OneAdvanced AI Admin license.

This article will show you, as an admin user, how to configure your organisation's guardrails. Guardrails are used to restrict the subjects that OneAdvanced AI will be allowed to engage with. The categories are taken from the ML Commons taxonomy of AI Hazards. To configure your organisation's guardrails;

  1. Navigate to https://myworkplace.oneadvanced.com.
  2. Within the AI menu (next to Files where Organisation spaces are managed) there is a new menu item: “Configurations”

Within here you can configure which Guardrails are enabled for your organisation.

  1. The Guardrails are all on by default in block mode. To allow a specific category, use the selector on the right-hand side and set it to disabled. This will disable the block.

By disabling these guardrails, users may encounter material that is not suitable for their needs or sensitivities. It is essential to carefully consider the potential consequences of removing these protections and to ensure that users are aware of the potential risks involved.

We would only advise removing these if content required for your line of business is being blocked.
  1. Once configurations have been set, select 'Save Changes'. Any changes configured will remove the Guardrail for that category.
The AI model itself has internal guards which cannot be controlled through configuration. The difference you will see is instead of the response stating the request was blocked due to OneAdvanced policies. The response will be along the lines of “I cannot help you with that”.

How did we do?

Organisation Space

Contact