Views:

For data sovereignty, you can keep the evaluation of your AI applications under your own control.

This page describes the full self-hosted deployment: you deploy the AI scan service, including the judge, in your own AWS environment with a CloudFormation template, and TrendAI Vision One™ receives only the scan results. That upload can also be disabled.
For many scenarios, a lighter option is enough, such as pointing AI Scanner at a custom judge model. You can configure any provider that supports the judge role (OpenAI-compatible, Amazon Bedrock, Azure OpenAI, or Google Vertex AI) in minutes with no AWS deployment, in the interactive wizard or in the scan configuration file. For more information, refer to AI Scanner configuration examples by AI endpoint provider. Use the full deployment below when you need the judge to run inside your own infrastructure.
For the TrendAI™-hosted configuration, refer to Configure TrendAI™-hosted AI scan settings.

Before you begin

  • A TrendAI Vision One™ API key with AI Scanner permissions.
  • The TrendAI™ Artifact Scanner (formerly TMAS) CLI, installed and on your path. For more information, refer to Configure TrendAI™-hosted AI scan settings.
  • An AWS account with permission to deploy CloudFormation stacks, or an existing TrendAI Vision One™-managed AWS account.
The same key authenticates the self-hosted scan service and the upload of results.

Part 1: Deploy the AI scan service in AWS

To configure a new AWS account

  1. In TrendAI Vision One™, go to Cloud Security → Cloud Accounts → AWS.
  2. Select Add Account.
  3. For Deployment Method, select CloudFormation.
  4. Select the account type:
    • Single AWS Account
    • AWS Organization: provide an Account name and Description to display in Cloud Accounts. Specify the Organizational Unit ID. When you add the AWS Organization, all member accounts without a specified alias in AWS receive an automatically generated name in Cloud Accounts.
  5. Select Next.
For more information, refer to Connect an AWS account using CloudFormation.

To update an existing AWS account

  1. Select Update AWS account, then select the account name.
  2. In Cloud Accounts Settings, select the Stack Update tab.
  3. Select the AWS region for CloudFormation template deployment.
  4. Select the Server & Workload Protection instance to associate with the account.
  5. To add custom tags to the resources deployed by TrendAI Vision One™, select Resource tagging and specify the key-value pairs. Select Create a new tag to add up to three tags.
  6. Select Next.
For more information, refer to AWS Stack update.

Enable AI Application Security

  1. Enable AI Application Security.
  2. Select the Deployment.
  3. Select Enable AI Scanner.
  4. In a new tab in the same browser session, sign in to the AWS account you want to connect using a role that has administrator privileges.
  5. For a new AWS account, select Next.
  6. To use automated deployment:
    1. Select Automated for the deployment type.
    2. Select Launch Stack to launch the CloudFormation template in the AWS console.
    3. Complete the steps in Quick Create Stack.
  7. To use manual deployment:
    1. Select Manual for the deployment type.
    2. Select Download the template and stack parameters as a .zip.
    3. Go to Stacks under CloudFormation.
    4. Create a new stack using the parameters in the .zip file.
  8. For an existing AWS account:
    1. Under Update the CloudFormation template, select Copy S3 URL.
    2. To review the template before deploying, select Download and Review Template.
    3. Go to Stacks and select the stack name (Vision-One-Cloud-Account-Management).
    4. Select Update.
    5. Select Replace current template.
    6. Paste the template S3 URL into Amazon S3 URL.
    7. Select Next.

Part 2: Get the self-hosted endpoint

  1. In the AWS management console, open the CloudFormation stack, then select the Outputs tab and copy the ScannerAPIEndpoint. You need this endpoint to point AI Scanner at your self-hosted instance.
  2. In TrendAI Vision One™, select Done.

Part 3: Run a scan against the self-hosted service

Start the scan the same way as a TrendAI™-hosted scan, and point the judge at the endpoint from Part 2. The self-hosted service authenticates with your TrendAI Vision One™ API key. No separate judge credential is needed.

Interactive wizard

  1. Start the wizard:
    ./tmas aiscan llm -i
  2. Configure the target and attacks as usual. For more information, refer to Configure TrendAI™-hosted AI scan settings.
  3. In Advanced Settings, open the Judge config section and select Custom Judge endpoint. Enter the ScannerAPIEndpoint from Part 2. Leave the API key blank to use your TMAS_API_KEY.
  4. Complete the remaining sections and start the scan.

Configuration file

In your scan configuration file, add a judge block that points at the self-hosted endpoint. The model field is required by the schema but is overridden by the self-hosted service:
version: 2.11.0
name: Self-Hosted Judge Scan
description: Scan with the judge running in my AWS environment
target:
  name: my-ai-app
  endpoint: https://api.openai.com/v1/
  api_key_env: TARGET_API_KEY
  openai:
    model: gpt-4o
judge:
  endpoint: https://<api_gw>.execute-api.us-east-1.amazonaws.com/prod/judge
  api_key_env: TMAS_API_KEY   # your TrendAI Vision One™ API key authenticates the service
  openai:
    model: trend-judge        # required by the schema; overridden by the service
attack_objectives:
  - name: Sensitive Data Disclosure
output:
  - local:
      format: html
      filename: results.html
      chat_history: true
Run the scan with the configuration file:
./tmas aiscan llm -c config.yaml

Results and data privacy

  • Chat transcripts are excluded by default when a custom judge, including the self-hosted service, scores the scan, to protect data privacy. You can control transcript inclusion per destination (TrendAI Vision One™ upload, each local report) in the wizard's output settings or with the chat_history field. TrendAI™ Artifact Scanner v2.296.0 or later is required to control inclusion per destination.
  • The TrendAI Vision One™ upload is optional. To prevent any scan data from being sent to TrendAI Vision One™, disable the TrendAI Vision One™ destination in the output settings, or list only local destinations in the output block of your configuration file. Results then appear only in your local reports.
  • Personally identifiable information (PII) redaction applies as with any scan: detected personal data is replaced with labeled placeholders in results before they are stored or uploaded.