AI-SPM
AI-SPM (Artificial Intelligence Security Posture Management) is a comprehensive approach to maintaining the security and integrity of AI/ML systems. It involves continuous monitoring, assessment, and improvement of the security posture of AI models, data, and infrastructure. It extends beyond traditional security to address risks unique to AI, such as data poisoning, model theft, and adversarial attacks.
The rapid adoption of AI has expanded the attack surface, introducing new risks in many traditional security domains. Use Cloud Exposure AI-SPM to protect each of these areas.
What AI Resources and Technologies Does Tenable Detect?
Cloud Exposure detects the following AI-related resources:
- AI workloads — Cloud native AI services detected via traditional cloud scanning or considered unmanaged (shadow AI).
- AI software — AI-related libraries that are installed on workloads, which are detected via workload scanning. For example, PyTorch, TensorFlow, and LlamaIndex.
- Data repositories — Repositories containing sensitive assets such as training datasets, system prompts, and uploaded files used in retrieval.
AI-SPM Use Cases by Domain
The following AI-SPM use cases demonstrate how the implementation of AI has expanded the traditional attack surface.
| Domain | Use Case (Risk) |
|---|---|
| Data Protection |
|
| CSPM |
|
| Workload Protection |
|
| Identity |
|
Sample AI-SPM Attack Path Flow
The following sample attack path flow describes how Cloud Exposure detects an overpermissive AWS Bedrock pipeline that is trained on sensitive data.
- Cloud Exposure analyzes IAM permissions for a user with access to a Bedrock pipeline and then utilizes CIEM capabilities to:
- Identify excessive permissions that can be removed.
- Detect risks from third-party vendors, and more.
- The user accesses the AWS Bedrock Pipeline (composed of an Agent and a Custom Model) via the identified IAM permissions.
- Tenable analyzes the pipeline's configurations to ensure secure deployment and compliance with best practices for AI workloads in AWS, leveraging CSPM capabilities.
- The Custom Model uses Stored Training Data from an S3 Bucket (containing file types like CSV, JSON, and TXT).
- Tenable discovers data used or generated by AI models, classifying its sensitivity, and assessing access paths and storage locations to prevent exposure or unauthorized access to sensitive information.
