What's New

This release rebuilds Detect AI around a two-phase workflow that separates finding your repositories from scanning them for AI, with supporting improvements to Bills of Materials and the public API.

Two-Phase Discovery and Detection

Detect AI now runs as two independent stages. Discovery enumerates the repositories in a version control system connection, and detection scans those repositories for AI. Previously the two were coupled, so a failure in discovery blocked detection and neither could be retried on its own. For connections with thousands of repositories, the entire pipeline had to finish before any results were available.

The two stages now trigger, monitor, and retry independently. Discovery results are stored before detection begins, and detection runs against that stored repository list. A discovery run can complete without triggering detection, and detection can run against a previously discovered list whenever you are ready. This makes scanning at enterprise scale faster and more resilient, because a problem in one stage no longer forces you to restart the other.

Targeted Scanning

You now control how much of your environment a scan covers. Detection can be triggered at three levels: every repository in an integration, every repository in one or more selected projects, or one or more specific repositories. A confirmation dialog shows the scope before the scan runs.

Targeted scanning removes the need to reprocess an entire connection when only a subset of repositories has changed. You can scan just the repositories you are responsible for, or just those with recent activity, which reduces both cost and turnaround time for large connections.

Bulk Actions

The Detect AI page supports acting on many items at once. Select multiple projects or repositories to rediscover their repositories, run detection, or create Bills of Materials in bulk from a single actions bar. The Detect AI Everywhere action scans every repository across your connections in one step.

When you create Bills of Materials from a mixed selection, only the repositories where AI was detected produce a BOM. Bulk-created BOMs appear in the Bill of Materials table as their individual scans finish.

Create BOMs from Detect AI Results

You can now generate a Bill of Materials directly from a discovered repository without leaving the Detect AI page. Create a BOM from a repository's row action, from the repository detail panel, or in bulk from a multi-selection. This closes the gap between finding AI in a repository and documenting it, so analysts triaging results no longer have to navigate away to start a BOM scan.

New Integrations and Detect AI Pages

Integrations and Detect AI each have their own section in the left navigation. Integrations is now a tabbed page covering VCS and Cloud Sensor connections, with discovery status shown directly in the table. Detect AI is a dedicated page with separate views for discovered projects and repositories, giving you room to sort, filter, and act on results.

Language & Version on the BOM Technology Tab

The Technologies tab of a Bill of Materials now includes Language and Version columns, along with a Language filter. When the same package name appears in more than one language, the tab shows each language, so a package like requests in Python is no longer indistinguishable from a same-named package in another language. The Version column shows the resolved version from the lockfile or manifest.

Public API: API Systems Endpoint

The public API adds a read-only, cursor-paginated endpoint for listing a tenant's AI Systems, consistent with the existing endpoints for Bills of Materials, vulnerabilities, and Arena models. Enterprise customers can now stream their AI System inventory into governance and GRC tools using the same pagination pattern as the other public endpoints.

Clearer BOM Category Definitions

The descriptions for the Models and Technologies categories on the Bill of Materials detail page have been rewritten for clarity. The underlying detection logic is unchanged; only the descriptive text was updated.

Did this answer your question?