0
2.2.3
Australia, Zurich Patch 4, Zurich
Standalone Application
Vulnerability management teams can use AI Security Exposure Management to help them import various types of security exposures in the AI assets including AI model vulnerabilities, AI model validation (automated red teaming) alerts, and AI posture findings (configuration issues in AI agents or tools, etc.) and manage these issues along with other types of exposures in the Unified Security Exposure Management workspace.
The AI Exposures dashboard provides you with a view into the critical security vulnerabilities of your AI attack surface. You have the option to use a generative AI skill to help you determine if any of the threats might be already mitigated and help you prioritize high risk exposures and defer lower risk exposures that have mitigations or guardrails already in place.
- Filter three categories of AI exposures that are displayed on a dedicated dashboard in the AI Exposures module in the Unified Security Exposure Management workspace:
- AI vulnerabilities - Vulnerabilities discovered in open source AI models that are published in repositories. These vulnerabilities are detected by third-party AI security tools and imported into AI security exposure management.
- AI validation findings - Findings from third-party automated penetration testing or automated red teaming done to verify the behavior of some of these models.
- AI posture findings - Configuration risks identified in AI agents, tools, instruction prompts etc. by third-party AI security tools in platforms such as Microsoft Copilot Studio, AWS, Google Vertex AI etc.
- Use automated response workflows for prioritization and remediation of AI exposures.
- Supports AI security platform integrations: Cisco AI Defense and Palo Alto Networks Prisma AIRS.
- For customers using Now Assist for Vulnerability response:
- Ingest data related to runtime guardrails enabled in AI security platforms.
- Use a Now Assist skill to help you determine if any of the model validation findings are mitigated by these guardrails. This generative AI skill can help explain why a guardrail helps mitigate a threat for a particular validation finding.
- Use an AI agent to help you defer and request exceptions for validation findings with eligible guardrails already in place as identified by the Now Assist skill. You can also chat with the AI agent to help you create exception rules to automatically defer similar findings in the future.
- Supported AI security platforms for this functionality: Cisco AI Defense.
New:
- Employee-facing remediation tasks for AI posture findings. Remediation tasks are now generated for AI posture findings and assigned directly to the agent owner, who can resolve issues or request exceptions from Employee Center without security role access. Task outcomes update the posture finding state, and closure is verified on the next integration run. This is useful for assigning the issues on agents created by business users in platforms such as Copilot Studio.
- Handling of asset owner resolution during task creation. The employee task creation process synchronously resolves asset owners, eliminating race conditions between asynchronous owner lookup and task creation. Tasks are reliably created for eligible posture findings referencing newly discovered assets.
- Configurable assignment rules for posture findings. Admins can now configure assignment rules to route posture findings to a user reference field (such as agent owner) on the discovered AI asset, rather than only to a group.
- Link discovered asset to AI assets. Host information such as VM, container, or git repo is now linked from discovered assets to AI assets, providing additional context for posture findings. This is for findings on agents or MCP servers found running in hosts such as EC2 or containers in cloud.
- UI action for creating AI Exposure Tasks. Vulnerability analysts can now trigger a "Create employee" action on AISEM posture findings to assign a task an end user for a configuration issue.
- Configurable property in advanced settings. A new "AI Security Exposure Management" tab is available in advanced settings, allowing admins to enable or disable employee task creation for AI posture findings.
Changed:
- Enhanced exception approval table name construction. Posture finding approval records now have improved naming logic for exception approvals.
- Updated handler for ingesting discovered asset references. The handler now ingests discovered asset sys IDs and populates them across AI assets, ensuring host information is accurately linked.
- Updated UI configuration and unit tests for employee tasks. The employee task UI has been updated and tested for improved usability and reliability.
Not applicable for this application version.
Required applications:
- Unified Security Exposure Management (sn_vul) v30.7.7
- If you want to use the AI guardrails helper skills you must install the Now Assist for Vulnerability Response application (sn_vul_ai).
- AI Security Common (sn_sec_ai_cmn) v1.1.1
- Configuration Compliance (sn_vulc) v30.6.7
Supported family releases:
- Brazil
- Australia
- Zurich
Supported integrations:
- The Prisma AIRS Integration for AI Security Exposure Management, a third-party integration supported by AI Security Exposure Management to import model scan results, model configuration rule violations, and model validation/attack data for AI models.
- The Palo Alto Prisma AIRS Service Graph Connector imports AI inventory data from Palo Alto Prisma AIRS and populates the ServiceNow CMDB.
- The Cisco AI Defense application is a third-party integration supported by AI Security Exposure Management to import AI Defense scan results and AI model validation data for AI and machine learning models.
- The HiddenLayer Service Graph Connector and the HiddenLayer Integration for AI Security Exposure Management are third-party integrations supported by AI Security Exposure Management to import AI model vulnerability data.