See the sections below for frequently asked questions about Agent Risk Manager.

What is the difference between the Early Adopter program and Early Access?

Early Adopter Program

The Early Adopter program allows customers to register their interest in using Agent Risk Manager. Through this program, KnowBe4 will keep you updated on Agent Risk Manager features and releases.

Early Access

Early Access is a limited-time preview of the product, designed to gather feedback from a small test group. If you meet the eligibility criteria, Agent Risk Manager should already be enabled for your account: see the quickstart guide for steps to access the product.

How can I leave feedback?

We highly recommend joining our KnowBe4 Community's Agent Risk Manager Early Access Agent Risk Manager Early Access (link opens in new window) group to give feedback and discuss product features with others. For more information, see our KnowBe4 Community Guide.

Is there a video walkthrough?

No videos or walkthroughs are planned for this stage of development. Any videos would quickly become outdated as we continuously grow and improve Agent Risk Manager.

Can admins see prompt text in the audit log?

Only after sensitive information has been automatically redacted can admins review prompt text in the audit log.

What AI integrations are planned?

For a list of currently supported AI vendor integrations, see our quickstart guide.

We plan to add the following direct AI vendor integrations in the future:

• Google Gemini
• OpenAI ChatGPT

How is my data handled?

Data Collection

When Agent Risk Manager evaluates an AI agent's tool execution, it receives the user's prompt, the tool inputs, and the surrounding conversation context. These user prompts and tool inputs are needed for the detection engines to detect risky AI interactions. The conversation context is also included so that the detection engines can evaluate the prompt against what came beforehand and better understand the context.

Data Processing

All analysis runs on models hosted by KnowBe4 in KnowBe4-controlled AWS infrastructure.

When sensitive information is detected in a prompt, it is masked before the record is kept. Agent Risk Manager keeps only the detection, not the sensitive information.

Data Storage

Data is encrypted at rest and in transit, and prompt content is never sent to any third-party AI providers.

KnowBe4 Data Practices

For detailed information on how KnowBe4 collects and secures your data, see our KnowBe4 Security articles and KnowBe4 Data Usage & Artificial Intelligence.