logo
開発者ドキュメント
OverviewUser GuideAPI ReferenceBest PracticeCHANGELOG
検索
日本語
User Guide / Agent / Security Mechanism
Security Mechanism
最新の更新：17 days ago

Security Mechanism

GPTBots Agents support the following two security mechanisms to ensure compliance and protect user privacy.

Moderation

Moderation performs multi-dimensional content moderation on both user inputs and agent-generated outputs. This includes, but is not limited to, detection of violence, sexually explicit material, politically sensitive topics, and other harmful content.

Currently, moderation is supported for text and image content; other multimedia types are not yet supported.

If a user's input fails the content moderation check, it will not be submitted to the agent. Likewise, if the agent's output fails moderation, it will not be displayed to the user.

Enable/Disable

In the "Security Mechanisms" section of the agent configuration page, you can enable or disable the moderation feature.

Config

  • Service: Select the content moderation service to use. Currently supported options include OpenAI Moderation, Keyword, and Custom API Extension.
  • Keywords: Required only when Keyword is selected as the service. You can define a list of keywords to be flagged during moderation.
  • Moderation: Choose whether to moderate INPUT (user-submitted content), OUTPUT (agent-generated content), or both.
    • Type: Select the type of content to be moderated. Currently, text and image are supported.
    • Preset Response: The response the agent will return to the user if the content fails moderation.

Data Anonymization

User input is anonymized to remove any personally identifiable information (PII) before being submitted to the agent. After the agent generates a response, the original PII is restored in the output.

This approach ensures that large language model providers do not have access to users' personal information, while still allowing the agent to function effectively.

Users can define and select which types of PII should be anonymized.

Enable/Disable

In the "Security Mechanisms" section of the agent configuration page, you can enable or disable the data anonymization feature.

Config

  • Method: Currently, only the replacement method is supported. Identified entities are replaced with placeholder tags. For example, John Doe would be replaced with <PERSON_1>.
  • Entity Groups: Select the entity groups to be anonymized. When the system detects entities belonging to the selected groups, they will be anonymized accordingly.