AI Defense Updates

Support for Azure AI Services

This release introduces AI asset visibility support for Azure AI Services. AI Defense can now validate Azure OpenAI-hosted models, building on existing support for AWS Bedrock-hosted models. This includes new support for asset types in Azure, including deployments, assistants, vector stores, and data files.

Validation

AI Defense validation offers number of improvements in this release:

• API-based testing: Launch and manage validation runs from the new AI Defense Management API (customer login required)

• New support for validation of guardrail-protected enforcement points

• Broader test coverage, now applying an expanded prompt bank with over 50 techniques and over 100 intents (up by 30 and 100 percent, respectively)

Runtime Protection

AI Defense runtime protection adds the following in this release:

• New Management API for setting up runtime protection: Integrate with your CI/CD pipeline to register and policy-protect your AI applications with AI Defense: https://developer.cisco.com/docs/ai-defense-management/ (customer login required)

• SDK: New Python SDK for the AI Defense Inspection API: https://github.com/cisco-ai-defense/ai-defense-python-sdk

Platform Security: User Activity Log

The new AI Defense user activity log tracks user interactions within AI Defense. The activity log is accessible to AI Defense administrators via the Administration panel. This feature ensures accountability and supports regulatory compliance by tracking all changes, allowing identification of unauthorized or erroneous modifications.