Skip to main content
September 5, 2026 This release expands Roe’s fraud monitoring and investigation workflows, adds post-action automation after agent runs, and brings the first scheduled external ingestion source online. It also improves livestream evidence capture, agent metrics defaults, and workflow reliability under load.

What’s new

  • Agents Fraud Investigation agents now monitor recurring risk patterns with incremental registry updates and in-investigation pattern search.
  • Agents Fraud and AML agent creation now uses a guided four-step setup wizard.
  • Agents Whatnot monitoring now resolves live status from fresh web extraction instead of cached profile data.
  • Agents Whatnot monitoring now captures live chat during frame capture for continuous evidence.
  • Agents Whatnot monitoring now supports optional live audio capture with transcription.
  • Agents Whatnot monitoring now captures live auction prices and fixes black-video sessions.
  • Agents Livestream capture now aligns frames, speech, and chat on one shared timeline.
  • Agents Agent post actions now run connector write operations after successful jobs and support retries.
  • Integrations Sift is now available as a browser context source with read-only review overview and network tools.
  • Integrations URLHaus is now live as Roe’s first scheduled external ingestion source.
  • SOPs Policies now support severity bands and ordered verdict overrides across authoring, runtime, and diffs.
  • QA Fraud indicators now abstain when recurrence comes from one applicant resubmitting.
  • QA Fraud indicators now keep only their own supporting documents and previews.
  • QA Workflow cancellations reported by callbacks now stay marked CANCELLED instead of FAILED.
  • QA Workflow startup now handles large job bursts more reliably with stricter capacity checks and retry hardening.
  • QA SageMaker-backed URL finding and web extraction now use longer activity timeouts.
  • QA SerpAPI text search now runs in its own worker activity instead of blocking other agent work.
  • Dashboard Agent Metrics now defaults to all versions across the agent’s full job history.
  • Dashboard Fraud risk registry now shows promoted and declined indicators in one unified view.
  • Dashboard Fraud indicator pages now include case dossiers, disposition mix, and paginated registry tables.
  • Dashboard Address-based fraud indicators now embed Google Maps directly on the indicator page.
  • Dashboard SQL result tables now expand into a full-size dialog with sticky headers.
  • Security Security updates resolved open dependency advisories across backend, frontend, and agent services.