Model Security Leaderboard

How secure is your AI model? Ranked by real adversarial testing — not benchmarks.

—
Models Tested
—
Attack Modules
—
Total Probes
14
Security Domains
Loading leaderboard data...

Get Your Model's Risk Snapshot

Free one-page report with breach rate, risk tier, and top 3 vulnerable categories.

We'll email your report within 24 hours.
Methodology

How We Score

Transparent, defensible, weighted by real-world consequence.

1.0x
Text Attacks
Prompt injection, jailbreaking, encoding bypass. Reputational impact.
1.5x
Agent Attacks
Tool calling, action injection, cross-agent contamination. Operational impact.
2.0x
Infrastructure
MCP exploitation, containment escape, supply chain. Systemic impact.
3.0x
Safety-Critical
OT/ICS, autonomous vehicles, robotics. Kinetic impact.

CLS Risk Score = Weighted Breach Rate × (1 + Severity Modifier)
Three independent AI judges cross-validate every finding. Coverage tier shows testing depth. All weights published.

Full Methodology →