Goal Summary:

RHACS can ingest and scan AI Bill of Materials (AIBOMs) in SPDX 3.0 and CycloneDX 1.6 formats from external sources such as OCI registries, model registries, and CI/CD pipelines. This allows security and platform teams to identify vulnerabilities in AI/ML workloads, gain visibility into deployed models, and assess associated risk, extending RHACS’s SBOM-based security capabilities to AI workloads.

The focus here would be to deliver a Technology Preview offering that is functional and allows AI-BOMs to be ingested and vulnerability reports created based on the ingested AI-BOMs. We look to take customer feedback and improve on the implementation for GA

Goals and expected user outcomes:

• Users can ingest AIBOMs into RHACS in SPDX 3.0 or CycloneDX 1.6 formats.

• RHACS scans all components listed in the AIBOM for vulnerabilities using the existing Scanner V4/Clair workflow.

• Vulnerability reports are generated in the same way as for SBOMs.

• Metadata from the AIBOM, including component name, version, hash, supplier, and dependencies, is preserved for traceability and audit purposes. 

Acceptance Criteria:

• RHACS can ingest AIBOMs in SPDX 3.0 and CycloneDX 1.6 formats from external sources (OCI registries, model registries, CI/CD pipelines).

• All ingested components are scanned for vulnerabilities using Scanner V4/Clair.

• Vulnerability scanning produces reports in the same format as SBOM vulnerability reports.

• AIBOM metadata is preserved during ingestion and available via UI and API.

• Ingestion and scanning performance remain within current SBOM SLAs.

Success Criteria or KPIs measured:

• Amplitude statistics capturing how many AIBOMs were ingested and scanned for vulnerabilities.
• Number of actionable vulnerability reports generated for AI workloads.

Use Cases (Optional):

• A model registry publishes a CycloneDX AIBOM for a deployed AI model; RHACS ingests and scans it, generating a vulnerability report.

• CI pipelines produce SPDX 3.0 AIBOMs for AI/ML services; RHACS scans them and includes results in the standard vulnerability dashboard.

• Security engineers query RHACS to view vulnerability reports for all AI workloads in a cluster.

• Auditors retrieve vulnerability reports for AI workloads for compliance and risk assessment purposes.

Out of Scope (Optional):

High-level list of items that are out of scope. Initial completion during Refinement status.

<your text here>