Skip to main content

Dashboard

Summary Tiles​

Three tiles across the top of the page give a quick read on feed coverage:

  • Total Threat Reports — total count of disclosed vulnerability reports submitted by security researchers, with a sparkline showing the trend over time.
  • Most Affected Vendor — the vendor with the highest share of vulnerability reports, shown as a percentage alongside the vendor name.
  • Scanner Probe Available — percentage of vulnerability reports that have an active scanner probe for automated detection.

Each tile is clickable and jumps to the corresponding section or filtered view.

Visualizations​

Below the tiles, the dashboard renders four data visualizations:

  • Most Affected Models — top AI models ranked by percentage of disclosed vulnerabilities affecting them. See LLM Model Cards for per-model security profiles.
  • Research Origins — geographic distribution of vulnerability reports based on the researcher's location.
  • Jailbreak Taxonomy: Category — distribution of vulnerabilities across jailbreak taxonomy categories, showing which attack types are most prevalent.
  • Cross-Model Vulnerability — cumulative count of vulnerabilities that affect multiple AI models, indicating widespread security issues.

Latest Reports​

A table on the right surfaces the most recently disclosed reports. Each row links to the full report and shows:

  • Title of the disclosure
  • Max Score — highest test result recorded against any model
  • Models Affected — visual indicator of how many models the report applies to (capped at 10)
  • Status — current publication state of the disclosure

An All reports button opens the full Reports list.

Vendors and Techniques Heatmap​

At the bottom of the dashboard, a heatmap cross-references AI vendors with jailbreak techniques. Cell intensity indicates how frequently a given technique has been observed against a given vendor — useful for spotting which techniques generalize across the ecosystem versus which are vendor-specific.